Legged robot

Abstract

The application discloses a kind of foot type robots.Therein, including: at least one memory is encoded with instruction, instruction when by at least one processor executes and makes foot type robot: response received remote control equipment sends fight instruction as first fight instruction, after obtaining the initial spatial position of fight object, hit rate evaluation is carried out to multiple attack orientations by pre-computation;If the first target orientation that makes the probability of attack action hit the fight object in initial spatial position be greater than or equal to probability threshold value is screened out in the initial position of foot type robot, the orientation of the front surface of torso is adjusted from initial orientation to first target orientation in initial position;In the process of being adjusted from initial orientation to first target orientation, the current spatial position of fight object is continuously identified, and first update target orientation is continuously screened, and attack action is executed in the direction of current spatial position in response to the orientation of the front surface of torso being first update target orientation.

Description

Technical Field

[0001] This application relates to the field of robotics, and more specifically, to a legged robot. Background Technology

[0002] Currently, in combat scenarios involving legged robots, after a user issues a combat command via a remote control device, the legged robot often relies solely on the user to adjust its orientation and mechanically execute combat actions according to a preset trajectory. However, this method frequently results in low accuracy issues such as "missed punches" and "missed hits," leading to a poor user experience. Therefore, the technical problem of poor combat effectiveness for legged robots remains.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a legged robot to at least solve the technical problem of poor combat effectiveness of legged robots.

[0005] According to one aspect of the embodiments of this application, a legged robot is provided, comprising: a torso; a plurality of target fighting limbs and corresponding plurality of fighting limb ends respectively connected to the torso; at least one visual perception module configured to acquire visual information of the surrounding environment of the legged robot, the visual information of the surrounding environment including at least one fighting object; at least one communication module configured to receive a movement command or a fighting command sent by a remote control device; at least one processor; and at least one memory, the at least one memory encoding instructions, the instructions, when executed by the at least one processor, causing the legged robot to: move in a target direction indicated by the received movement command sent by the remote control device; respond to the received fighting command sent by the remote control device as a first fighting command, the first fighting command including target fighting action information; after acquiring the initial spatial position of the fighting object, perform a hit rate assessment on multiple attack directions through pre-calculation; if a target is selected from the multiple attack directions, at the initial position of the legged robot... If the probability of an attack hitting a combatant at the initial spatial position is greater than or equal to a probability threshold, then the orientation of the torso's forward face is adjusted from the initial orientation to the first target orientation. During this adjustment, the current spatial position of the combatant is continuously identified, and a new target orientation is continuously selected that makes the probability of the attack hitting the combatant at the current spatial position greater than or equal to a probability threshold. In response to the orientation of the torso's forward face being the new target orientation, the attack action indicated by the first combat command is executed in the direction of the current spatial position. The initial spatial position represents the spatial position of multiple points on the surface of the combatant's geometric bounding body in the world coordinate system when a combat command is received from the remote control device. The current spatial position represents the current position of multiple points on the surface of the combatant's geometric bounding body in the world coordinate system. Attack actions include all attack actions except for pure area-of-effect attacks.

[0006] According to another aspect of the embodiments of this application, another legged robot is also provided, comprising: a torso; a plurality of target fighting limbs and corresponding plurality of fighting limb ends respectively connected to the torso; at least one visual perception module configured to acquire visual information of the surrounding environment of the legged robot, the visual information of the surrounding environment including at least one fighting object; at least one communication module configured to receive movement commands or fighting commands sent by a remote control device; at least one processor; and at least one memory encoded with instructions, which, when executed by the at least one processor, cause the legged robot to: move in the target direction indicated by the received movement command sent by the remote control device; and execute the pure range attack action indicated by the received fighting command sent by the remote control device, the second fighting command including target pure range attack action information.

[0007] According to another aspect of the embodiments of this application, another legged robot is also provided, comprising: a torso; a plurality of target fighting limbs and corresponding plurality of fighting limb ends respectively connected to the torso; at least one visual perception module configured to acquire visual information of the environment surrounding the legged robot, the visual information of the environment including at least one fighting object; at least one communication module configured to receive a movement command or a fighting command sent by a remote control device; at least one processor; and at least one memory encoded with instructions that, when executed by the at least one processor, cause the legged robot to: move in the target direction indicated by the received movement command sent by the remote control device; acquire the initial orientation of the forward face of the torso and determine the starting orientation of the fighting action indicated by the fighting command to be executed in response to the received fighting command sent by the remote control device; adjust the orientation of the forward face of the torso from the initial orientation to the starting orientation; and execute a fighting action in the fighting direction indicated by the fighting command in response to the orientation of the forward face of the torso being the starting orientation.

[0008] In this embodiment, if a movement command is received from a remote control device, the user can move in the target direction indicated by the movement command. If the received combat command from the remote control device is a first combat command, the initial spatial position of the combat target can be obtained. After obtaining the initial spatial position, the hit rate of various attack directions can be evaluated through calculation. If a first target direction is selected from the various attack directions, and the probability of the attack hitting the combat target in the initial spatial position is greater than or equal to a probability threshold, the orientation of the torso's forward face can be adjusted from the initial orientation to the first target orientation. During the process of adjusting the orientation of the torso's forward face from the initial orientation to the first target orientation, the current spatial position of the combat target can be continuously identified, and a first updated target orientation with a probability of the attack hitting the combat target in the current spatial position greater than or equal to a probability threshold can be continuously selected. If the orientation of the torso's forward face is the first updated target orientation, the attack action indicated by the first combat command can be executed in the direction of the current spatial position. In other words, in this embodiment, the legged robot's autonomous takeover effectively avoids the combat deviations caused by mechanically executing combat commands in related technologies. For the first combat command, the legged robot does not directly execute the action according to the user's command, but instead takes over the orientation adjustment process of the torso's forward face, executing attack actions other than pure range attacks. This takeover mechanism ensures that the legged robot's combat execution no longer relies solely on the user's precise aiming to improve hit rate (e.g., hitting the combat target), whereas related technologies lack the ability for legged robots to autonomously decide and adjust their orientation. This achieves the technical effect of improving the combat effectiveness of legged robots and solves the technical problem of poor combat effectiveness in legged robots. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0010] Figure 1 This is a schematic diagram of a legged robot according to an embodiment of this application;

[0011] Figure 2 This is a schematic diagram of the first side of a legged robot according to an embodiment of this application;

[0012] Figure 3 This is a schematic diagram illustrating the hit probability of a legged robot according to an embodiment of this application.

[0013] Figure 4 This is a schematic diagram illustrating the hit probability of another legged robot according to an embodiment of this application;

[0014] Figure 5 This is a schematic diagram illustrating the hit probability of another legged robot according to an embodiment of this application;

[0015] Figure 6 This is a schematic diagram of a fighting scene of a legged robot according to an embodiment of this application;

[0016] Figure 7 This is a schematic diagram of a fighting scene of another legged robot according to an embodiment of this application;

[0017] Figure 8 This is a schematic diagram of another legged robot according to an embodiment of this application;

[0018] Figure 9 This is a schematic diagram of another legged robot according to an embodiment of this application. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] According to an embodiment of this application, a legged robot embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0022] This application provides a legged robot. This legged robot can be used to provide intelligent combat and human-machine collaborative attack control functions for preset application scenarios. These preset application scenarios may include the following scenarios in the fields of robot competition and human-machine interactive entertainment: legged robot combat events, intelligent combat training platforms, immersive combat game terminals, robot combat entertainment systems, etc.

[0023] Based on this, this application takes the field of robot combat and competition as an example to illustrate the specific implementation of the above-mentioned legged robot (hereinafter referred to as robot).

[0024] Figure 1 This is a schematic diagram of a legged robot according to an embodiment of this application, as shown below. Figure 1 As shown, the legged robot 10 includes a torso 11; multiple target fighting limbs 12 and corresponding multiple fighting limb ends 13 connected to the torso; at least one visual perception module 14 configured to acquire visual information of the surrounding environment of the legged robot, the visual information of the surrounding environment including at least one fighting object; at least one communication module 15 configured to receive movement commands or fighting commands sent by a remote control device; at least one processor 16; and at least one memory 17, the at least one memory encoding instructions, which, when executed by at least one processor, cause the legged robot to: step S102, move in the target direction indicated by the received movement command sent by the remote control device; step S104, respond to the received fighting command sent by the remote control device as a first fighting command, the first fighting command including target fighting action information, and after acquiring the initial spatial position of the fighting object, perform a hit rate assessment on multiple attack directions through pre-calculation; step S106, if selected from multiple attack directions, in the legged robot The robot's initial position is such that the probability of the attack hitting the fighting object at the initial spatial position is greater than or equal to a probability threshold. Then, at the initial position, the orientation of the torso's forward face is adjusted from the initial orientation to the first target orientation. In step S108, during the adjustment of the torso's forward face from the initial orientation to the first target orientation, the robot continuously identifies the current spatial position of the fighting object and continuously filters for the first updated target orientation, ensuring that the probability of the attack hitting the fighting object at the current spatial position is greater than or equal to a probability threshold. In response to the torso's forward face being the first updated target orientation, the robot executes the attack action indicated by the first fighting command in the direction of the current spatial position. The initial spatial position represents the spatial position of multiple points on the surface of the fighting object's geometric bounding body in the world coordinate system when the fighting command sent by the remote control device is received. The current spatial position represents the current position of multiple points on the surface of the fighting object's geometric bounding body in the world coordinate system. The attack action includes attack actions other than pure range attack actions.

[0025] Optionally, the target fighting limb 12 can be a movable limb structure connected to the torso for performing fighting actions. The function of the target fighting limb is to transmit power and form a kinematic chain for attacking or defending actions. The target fighting limb can include, but is not limited to, upper limbs and lower limbs. For example, the upper limb can be a structural unit such as an arm (upper arm, forearm). The lower limb can be a structural unit such as a thigh, lower leg. Each target fighting limb can have multiple degrees of freedom, enabling joint rotation, extension, and swinging movements to support the execution of preset fighting action trajectories. The target fighting limb can selectively participate in fighting actions (e.g., attacking actions, defensive actions) according to the specific content of the fighting instructions, and the movement state of the target fighting limb can be coordinated and controlled by the processor based on visual information and instruction type.

[0026] Optionally, the combat limb end 13 can be an execution part on the target combat limb used to perform combat actions. The aforementioned combat limb end can be a part directly connected to the target combat limb, an integral part of the legged robot itself, and a part that needs to make physical contact with the target object during the execution of combat actions; that is, the target limb end, for example, the fist, foot, knee, or elbow of the legged robot. The fist can be used to perform combat actions such as punches (straight punches, hooks, etc.); the foot can be used to perform combat actions primarily based on leg power, such as sweeping kicks, side kicks, front kicks, spinning kicks, and downward kicks; the knee can be used to perform combat actions such as knee strikes (e.g., front knee strikes, side knee strikes) and knee thrusts; the elbow can be used to perform combat actions such as elbow strikes (e.g., horizontal elbow strikes, rear elbow strikes) and smashing elbow strikes. The aforementioned combat limb end can also be a non-legged robot component, held by the target limb end, and a part that comes into contact with the target object during the execution of combat actions, such as a combat weapon. The aforementioned combat implements can include knives, clubs, etc. The swing trajectory of the combat limbs can be spatially located and tracked by a processor based on combat commands and visual information. The spatial position and direction of the combat limbs directly determine the effectiveness of the combat action. Combat limbs can also be referred to as attack ends, execution endpoints, or contact points.

[0027] Optionally, the at least one visual perception module 14 can be used to obtain visual information from the environment surrounding the legged robot. The visual perception module can be an environmental perception component configured for the legged robot, used to acquire visual information about the environment in which the legged robot is located, to support the identification and localization of the position, posture, and motion state of the fighting object. The visual perception module may include at least one imaging device (e.g., a camera). The imaging device can be deployed on the structural surfaces of the legged robot, such as its head, chest, shoulders, back, or limbs. It should be noted that the deployment position of the imaging device can be flexibly configured according to the field of view coverage requirements to achieve multi-angle, all-round environmental perception and visual information acquisition for the legged robot. For example, the imaging device can be a monocular camera, a binocular camera, a wide-angle camera, an infrared camera, or a combination of multiple cameras. The visual information output by the visual perception module can form the basis for subsequent forward orientation adjustment, trajectory matching of fighting actions, and pre-calculation of attack probability.

[0028] The visual information collected by at least one visual perception module 14 can include at least one object in the environment and can be used to describe the spatial visual data of the environment surrounding the legged robot. The content of the visual information may include, but is not limited to, the situation of the environment in which the legged robot is located and the situation of the fighting object in the environment. The situation of the environment in which the legged robot is located can describe the spatial structure and boundary features of the environment in which the legged robot is currently located, including but not limited to: the outline of the fighting arena, the boundary facilities of the fighting arena (e.g., octagonal cage fence), the ground texture of the fighting arena, the lighting conditions of the fighting arena, and the positions of fixed obstacles on the fighting arena. The situation of the environment in which the legged robot is located is used to establish a scene reference frame in the world coordinate system, supporting the legged robot in locating its own posture and planning its motion path. The situation of the fighting object can be the real-time spatial state information of the fighting object in the environment, including the posture information of the fighting object in the world coordinate system, the orientation of the forward face of the torso, and the outline boundary, etc. The situation of the fighting object is used to determine the probability of the attack action performed by the legged robot hitting, supporting the adjustment of its own forward face orientation, swing trajectory matching, and attack timing decision-making.

[0029] It should be noted that the deployment method of the at least one visual perception module 14 in this embodiment is not fixed. One deployment method is to deploy at least one visual perception module 14 on a legged robot, for example, on the chest or head of the legged robot. Another deployment method is to deploy at least one visual perception module 14 in the fighting arena where the legged robot fights its opponent. In the first deployment method, the legged robot collects visual information itself. In the second deployment method, the acquired visual information needs to be sent to the legged robot (e.g., through a communication module that communicates with the visual perception module), or it can be sent to the user's remote control device. No specific restrictions are placed on the deployment location of the at least one visual perception module 14 or the method by which the legged robot acquires visual information; these can be flexibly adjusted according to the actual situation, as long as the robot can perceive its environment and make decisions about fighting actions.

[0030] Optionally, at least one communication module 15 can be an external signal receiving component configured on the legged robot, used to receive movement commands or fighting commands initiated by the remote control device, realizing two-way information interaction between the legged robot and the control terminal (the user controlling the legged robot using the remote control device). The communication module can support wireless or wired communication methods, including but not limited to radio frequency communication, Bluetooth communication, Wireless Fidelity (Wi-Fi) communication, or infrared communication. The function of the communication module is to completely and reliably transmit the movement commands or fighting commands issued by the remote control device to the processor, serving as the triggering basis for the legged robot's movement and fighting.

[0031] The aforementioned remote control device can be an external control terminal operated by a user, used to send movement and combat commands to the legged robot, enabling the expression of intent and guidance of the robot's behavior. The remote control device can be a handheld controller, equipped with physical buttons, joysticks, a touchscreen (virtual buttons), or a voice input device, supporting user input of movement and combat commands to meet their needs. The movement commands are control commands issued by the remote control device to control the legged robot's movement within its environment. The command content may include the user's desired direction of movement and movement speed, used to instruct the legged robot to adjust its position within the environment. The movement commands do not contain any offensive or defensive intent, but rather guide the legged robot to perform basic movement behaviors such as obstacle avoidance, approach, retreat, or posture preparation within the environment. Movement commands can be triggered by the user through joystick, directional keys, or touchscreen swipes. The aforementioned combat command can be a control command issued by a remote control device to trigger the legged robot to perform combat behavior. The content of the aforementioned combat command may include, but is not limited to, at least one of the following: the content of the combat action, the direction of the combat action, the combat target of the combat action, etc.

[0032] Optionally, the at least one processor 16 is used to receive and execute instructions encoded by memory to realize logical decision-making and real-time control of the robot's movement and fighting behavior. The processor receives movement instructions and fighting instructions from the communication module, combines the visual information output by the visual perception module, analyzes the initial orientation of the torso, the initial spatial position, the position of the fighting target, the type of attack action, and the defensive requirements, and generates control signals based on a preset algorithm to drive the torso to adjust its orientation, the target fighting limb to execute trajectory movement, and the fighting limb end to complete attack or defensive actions. The at least one memory 17 can be a non-volatile or volatile storage unit configured within the legged robot for storing program instructions and data.

[0033] In this embodiment of the application, at least one memory 17 is encoded with instructions that, when executed by at least one processor, cause the legged robot to perform step S102: in response to a movement instruction sent by a received remote control device, move in the target direction indicated by the movement instruction.

[0034] In this embodiment, if the legged robot receives a movement command from a remote control device, the processor can, based on the target direction and movement speed contained in the movement command and the current state of the legged robot, execute displacement control to move towards the specified target direction at the corresponding movement speed. This process does not rely on a fixed motion pattern, but rather flexibly selects whether to adjust the orientation of the torso's forward face based on the characteristics of the movement command and the state of the legged robot, to achieve efficient, accurate, and compliant motion response. The movement control flow of the legged robot according to the movement command can be divided into two execution paths: one is to directly execute movement without changing the orientation of the torso's forward face, and the other is to adjust the orientation of the torso's forward face before or during movement to match the target direction. The processor can dynamically determine and select which execution path to follow, ensuring that the legged robot maintains a good balance between motion freedom and control response in the environment.

[0035] Optionally, in the first execution path described above, if there is a certain angular deviation between the target direction indicated by the movement command and the current orientation of the forward face, but the angular deviation does not exceed the preset motion tolerance range indicated by the movement command, and the current legged robot is not in the preparation stage of an attack or defense action, the processor does not need to trigger the torso's yaw rotation relative to the orientation. Instead, it directly drives multiple target fighting limbs to move in coordination, using the initial orientation of the torso as a reference, to complete the displacement towards the target direction. For example, if the movement command is "move one step to the left laterally," the processor interprets the movement command as a lateral translation requirement. It does not need to change the orientation of the torso's forward face. It can use joint coordination control such as the left leg abduction and lifting, the right leg support and stabilization, and the hip joint lateral sliding to move the overall center of mass of the legged robot to the left to the target position (e.g., the first target position), while the orientation of the torso's forward face remains unchanged.

[0036] It should be noted that the above-described movement method in this application embodiment is applicable to scenarios such as short-distance rapid displacement, dodging sudden attacks, adjusting standing posture, or reserving posture freedom for subsequent commands. The effect is that the response speed is fast, the action continuity is high, and the movement delay caused by turning actions is avoided.

[0037] Optionally, in the second execution path described above, if the target direction indicated by the movement command deviates significantly from the initial orientation of the torso, or if the movement action indicated by the movement command is a preparatory step required for an upcoming combat action, the processor determines that the orientation of the torso's forward face can be adjusted. At this time, the processor can execute a command to drive the torso to yaw around the vertical axis, gradually adjusting the orientation of the torso's forward face from its initial orientation to a target orientation that matches the target direction (e.g., the forward face facing the target direction). This adjustment process can be performed synchronously with the movement action or in segments. For example, if the movement command indicates a long-distance movement, the processor can prioritize adjusting the torso's forward face orientation before initiating a straight-line gait. If the movement command indicates a short-to-medium distance movement, the processor can simultaneously perform the torso's forward face orientation adjustment and leg coordination, achieving a composite movement of turning while moving.

[0038] For example, when the movement command is "take two steps forward" and the current forward face is 60 degrees to the right, the processor can calculate that the target orientation of the torso is directly forward. It then controls the torso to slowly yaw forward, simultaneously moving the left leg forward and the right leg following, completing the displacement. This method ensures that the legged robot's swing trajectory aligns with the intended movement command. Figure 1 To.

[0039] In this embodiment, the two execution paths described above can be dynamically decided by the processor based on a preset rule base. This rule base may include: if the movement distance is less than a preset distance (e.g., 0.5 meters) and there is no imminent combat command, then direct movement without adjusting the orientation of the torso's forward face is preferred; if the movement distance is greater than the preset distance (e.g., 0.5 meters), or if a combat command will be executed after the movement is completed, then the orientation adjustment of the torso's forward face can be initiated.

[0040] In summary, the aforementioned motion control process achieves a balance between the mobility and control precision of legged robots by distinguishing between two execution paths: one that adjusts the orientation of the torso's forward face and the other that does not. When no orientation adjustment is needed, the robot can quickly respond to short-distance displacement requirements, improving the continuity of the combat rhythm. When orientation adjustment is required, the legged robot actively calibrates the orientation of its forward face, improving the accuracy of subsequent combat actions. This mechanism not only avoids the rigid "command-is-turn" control mode of legged robots but also endows them with the ability to autonomously select appropriate movement strategies in combat environments, significantly enhancing human-robot collaboration efficiency and the realism and intelligence of the combat experience.

[0041] In this embodiment of the application, at least one memory 17 is encoded with instructions that, when executed by at least one processor, cause the legged robot to perform step S104: responding to a received fighting instruction sent by a remote control device as a first fighting instruction, the first fighting instruction including target fighting action information, and after obtaining the initial spatial position of the fighting object, pre-calculating the hit rate assessment of multiple attack orientations.

[0042] The first combat command can be a control command sent by a remote control device to trigger the legged robot to perform non-global combat behavior. The command content of the first combat command can include target combat action information. This target combat action information can include a type identifier for attack actions other than purely area-of-effect attacks (i.e., directional attack actions). The combat target can be a physical entity in the combat scenario with the legged robot, possessing a body structure and motion characteristics recognizable by the visual perception module. The combat target can include, but is not limited to, other legged robots, bionic combat robots, or dynamic targets with humanoid forms, and can also include animals.

[0043] The aforementioned attack actions may include a first attack action and a second attack action. The first attack action may be an attack action executed according to a fixed trajectory. The second attack action may be an attack action executed according to a generalized trajectory. The generalized trajectory may be a trajectory obtained by adjusting a preset trajectory for executing the attack action based on the part of the body to be attacked and the real-time situation of the combatant (e.g., real-time position).

[0044] Optionally, the forward surface of the torso can refer to the front surface area of ​​the torso of the legged robot that faces the direction of movement or the main interaction direction. In simple terms, the forward surface can correspond to the plane where the front chest of the legged robot is located. Figure 2 This is a schematic diagram of the forward face of a legged robot according to an embodiment of this application, as shown below. Figure 2 The forward surface refers to the torso surface corresponding to the direction the chest of the legged robot faces when it is standing or walking normally. A forward direction vector can be extended from the center of the legged robot's chest towards the front of the torso. The surface where the legged robot's chest is located is the forward surface. This forward direction vector can be used to represent the direction the legged robot is facing. For example, if the legged robot's chest faces due north, then the forward surface of the torso faces due north.

[0045] The initial position can be the spatial coordinates of the legged robot's torso's center of mass in the world coordinate system at the moment the first combat command is received. This initial position can be acquired and analyzed by the visual perception module at the instant the first combat command is triggered; it does not change with subsequent movement and serves as the initial spatial parameter for determining the probability of an attack hit. Alternatively, the initial position can be the spatial coordinates of the combat target in the world coordinate system at the moment the first combat command is received. This initial position can be obtained through visual information recognition and pose estimation.

[0046] The aforementioned initial spatial position can be used to represent the spatial positions of multiple points on the surface of the geometrically bounding body of the fighting object in the world coordinate system when a fighting command is received from a remote control device. For example, the initial spatial position can be the spatial coordinates of points on the surface of the geometrically bounding body of the torso of a legged robot in the world coordinate system at the moment the fighting command is received. Here, the aforementioned geometrically bounding body can refer to a minimal convex polyhedron or convex hull that closely conforms to the surface of the three-dimensional model of the fighting object. In the spatial position representation, multiple points on the surface of the aforementioned geometrically bounding body constitute a set of discrete spatial contours of the fighting object in the world coordinate system. By defining the coordinates of the points on the surface of the aforementioned geometrically bounding body, the precise geometric shape and spatial occupancy of the fighting object at the instant the fighting command is received can be accurately described.

[0047] The aforementioned attack actions, excluding pure area-of-effect attacks, refer to localized fighting behaviors with clear spatial directionality, a range of action limited to the swing trajectory of a specific target limb, and an attack effectiveness highly dependent on the orientation of the torso's forward face and its relative position to the target. These attack actions require the legged robot to adjust the orientation of its torso's forward face to the first target orientation before execution, ensuring spatial overlap between the swing trajectory of the end of the fighting limb and the target's body / limb; otherwise, the attack will miss. The range of action can be a limited local area, with directional constraints and no full-area coverage capability, and can include, but is not limited to, straight punches, left hooks, front kicks, elbow strikes, knee strikes, and forward thrusts. The fundamental difference between these attack actions and pure area-of-effect attacks is that successful execution requires calibration of the torso's forward face, relying on a spatial alignment mechanism. These attack actions can also be called directional fighting actions, where "directional" refers to orientation.

[0048] For example, the attack actions mentioned above, excluding pure area-of-effect attacks, can include directional attacks and directional area attacks. Directional attacks can refer to a legged robot executing a fighting action with a clear attack axis, pointing in a specific direction. Alternatively, it can refer to a legged robot executing a fighting action with a clear attack fan-shaped area, directed towards a specific narrow space. For example, directional attacks could include a straight punch, a front kick, or a downward elbow strike. Directional area attacks can refer to a legged robot executing a fighting action covering a certain spatial volume or fan-shaped area while maintaining a specific orientation. Compared to directional attacks, directional area attacks have a wider attack coverage, potentially involving the coordinated movement of multiple target limbs or a large-scale swinging motion of a single target limb. For example, directional area attacks could include a right hook, double push punches, and a spinning kick.

[0049] In this embodiment, after the legged robot receives the first combat command from the remote control device, the initial spatial position of the combat target in the surrounding environment can be obtained. The hit rate can be evaluated by calculating the legged robot's ability to hit the combat target at its initial spatial position from various attack orientations.

[0050] Optionally, during the process of acquiring the initial spatial position of the fighting object, the legged robot captures visual information of its surroundings in real time through at least one visual perception module. Feature data of the fighting object can be extracted from this visual information. Based on the visual information, the processor establishes a world coordinate system and calculates the three-dimensional coordinates of multiple points on the geometrically bounding surface of the fighting object in the world coordinate system, thereby determining the initial spatial position of the fighting object.

[0051] Optionally, during the hit rate evaluation of multiple attack orientations, after obtaining the initial spatial position of the fighting target, the processor constructs a kinematic model of the attack action indicated by the target fighting action information contained in the first fighting command. The processor defines a set of attack orientations. This set of attack orientations may include multiple discrete or continuous attack orientations that the legged robot's torso may adjust to in its initial position. For each attack orientation, the processor pre-calculates whether the target attacking limb or the end of the fighting limb, under the constraint of that attack orientation, overlaps with the spatial region where the initial spatial position is located when it moves along a predetermined swing trajectory (e.g., a fixed trajectory, a generalized trajectory). By calculating the overlap probability or determining whether the conditions for hitting the head, torso, or limbs are met, the processor obtains the hit probability corresponding to each attack orientation. The processor compares the hit probabilities of each attack orientation and selects the attack orientation with the highest hit probability (greater than or equal to a preset probability threshold) as the final first target orientation.

[0052] In this embodiment, the method significantly improves the decision-making intelligence and attack success rate of the legged robot when executing the first combat command by introducing a pre-calculated hit rate evaluation mechanism based on multiple attack orientations. This mechanism avoids the phenomenon of missing attacks due to blindly executing attack commands, enabling the legged robot to actively seek the optimal attack posture before executing the action, effectively overcoming operational errors caused by the user's remote control commands not fully considering the relative spatial position. Simultaneously, by quantifying the hit probability of different orientations, the legged robot can dynamically adapt to changes in the initial position of the combat opponent, ensuring that in complex combat scenarios, the attack action can accurately target the effective hit area of ​​the target, thereby enhancing the sense of control in human-computer interaction and the competitive experience of the game.

[0053] In this embodiment of the application, at least one memory 17 is encoded with instructions that, when executed by at least one processor, cause the legged robot to perform step S106: if a first target orientation is selected from a variety of attack orientations, and the probability of the attack action hitting the fighting object in the initial spatial position is greater than or equal to a probability threshold at the initial position of the legged robot, then at the initial position, the orientation of the forward face of the torso is adjusted from the initial orientation to the first target orientation.

[0054] The initial spatial position refers to the spatial position of multiple points on the surface of the geometrically bounding body of the fighting object in the world coordinate system when a fighting command is received from the remote control device. Attack actions include all attack actions except for pure range attacks, and the first target orientation is the direction corresponding to the forward face of the torso at the start of the attack action. The initial position of the legged robot and the initial spatial position of the fighting object can both be in the world coordinate system, and the initial orientation of the forward face of the torso can also be in the world coordinate system.

[0055] Optionally, the aforementioned initial orientation may refer to the physical angular state of the forward face of the legged robot in the world coordinate system at the instant it receives the first combat command. This initial orientation can be used to reflect the initial spatial posture of the legged robot when deciding to execute an attack action.

[0056] Optionally, the probability threshold can be a preset minimum probability standard for determining whether an attack action is executable (hitting the combat target) under the initial position and initial spatial position configuration. This minimum probability standard can be set based on the type of attack action, the size of the combat target, and tactical requirements. This probability threshold can serve as a boundary between orientation adjustment and displacement decisions, ensuring that the legged robot executes the attack action when the probability of a hit is sufficient, avoiding response delays or energy waste due to low-probability actions. For example, the probability threshold can be preset to 95% or 100%, without specific limitations. The 95% threshold is chosen to ensure a high tolerance for error in attack action execution and improve the visual appeal of the combat process.

[0057] The aforementioned first target orientation may be the same as or different from the initial orientation. The aforementioned first target orientation can be the direction corresponding to the forward face of the torso at the start of the attack action, enabling the effective execution of the attack. For the case of effectively executing the attack action, for example, if the aforementioned fighting action is an attack action other than a pure area-of-effect attack, the aforementioned first target orientation can be the orientation of the torso's forward face, pre-calculated to ensure that the attack action indicated by the first fighting command (excluding pure area-of-effect attacks) has the feasibility of hitting the fighting target, given the torso's initial position.

[0058] The first target orientation can satisfy the following three conditions: the first condition is determined in response to the first combat command; the second condition is that the probability of the attack action hitting the combat target is greater than or equal to the probability threshold after pre-calculation; and the third condition is the orientation of the forward face of the torso corresponding to the initial action of the attack action.

[0059] Optionally, regarding the first condition (determined in response to the first combat command), the analysis is as follows: The first target orientation can be directly generated by the first combat command received by the legged robot, which serves as the trigger source for the attack action. After parsing the first combat command, the corresponding swing trajectory model and associated parameters can be retrieved from the action library based on the command content, thereby limiting the candidate range of the first target orientation. For example, when the first combat command is a right straight punch, the orientation of the first face matching the swing trajectory of the right straight punch can be considered, while the orientation logic of other attack actions (such as hook punches) is not allowed to be introduced. The first condition ensures that the first target orientation is aligned with the user's command intent of the first combat command, avoiding unauthorized changes to the attack intent and ensuring the predictability and consistency of the control intent.

[0060] Optionally, regarding the second condition mentioned above (that the probability of the attack hitting the target is greater than or equal to a probability threshold after pre-calculation), the analysis is as follows: The first target orientation is not arbitrary, but rather a specific orientation whose hit probability reaches or exceeds the combat threshold after pre-calculating the trajectory of the end of the fighting limb and the position of the fighting target. The pre-calculation process is based on the initial position and initial spatial position, calculating the intersection probability of the swing trajectory and the geometric enclosure of the fighting target under various attack orientations at that position, and finally selecting the attack orientation with the highest probability, which is greater than or equal to the aforementioned probability threshold. For example, if the swing trajectory penetrates the opponent's head under an orientation with a 25° right rotation relative to the torso, the hit probability is 0.96; while under orientations with a 15° or 35° right rotation relative to the torso, the probability is less than 0.65, then the final selected attack orientation is a 25° right rotation relative to the torso. In this embodiment of the application, the second condition mentioned above can ensure that the selected attack orientation has actual tactical effectiveness, rather than just being based on directional alignment, thus avoiding misjudgments of "seemingly aligned but actually missing the target," and improving the attack success rate and the decision-making reliability of the legged robot.

[0061] Optionally, regarding the third condition mentioned above (the orientation of the forward face of the torso corresponding to the initial action of the attack), the analysis is as follows: The first target orientation can spatially correspond to the initial action of the attack. That is, the first target orientation can be the orientation of the first face required by the starting point of the swing trajectory, rather than the orientation at any moment during the attack (e.g., at the moment of striking or in the recovery posture). The initial action is the first identifiable preparatory posture in the attack sequence. For example, the forward shoulder lean of a straight punch, the backward shift of the center of gravity and torso twist of a sweeping leg. The orientation of the first face when executing the initial action determines the initial direction of motion at the end of the attack, ensuring that the legged robot is in the correct pre-swing posture before executing the attack, so that the subsequent trajectory tracking has physical consistency. For example, if the initial posture of the sweeping leg action requires the torso to twist 45° to the left, then regardless of where the fighting opponent is on the right side, the first target orientation must be 45°, rather than a straight line towards the fighting opponent. In the embodiments of this application, the third condition mentioned above can effectively prevent the legged robot from confusing "facing the fighting object" with "facing at the start of the action", ensuring the executability of the attack action in the mechanical structure and the standardization of the action library.

[0062] In this embodiment, the complete definition of the first target orientation, constituted by the above three conditions, is a physical starting posture that can be triggered by user commands, has been pre-calculated and verified to have hit capability, and strictly matches the attack action. These three conditions together ensure that the legged robot, before executing the attack, not only conforms to the operational intent and possesses technical feasibility, but also maintains the standardization and consistency of the action design.

[0063] Optionally, in the process of selecting the first target orientation that meets the probability threshold for hitting the fighting target, the processor can evaluate multiple attack orientations one by one based on the acquired initial spatial position of the fighting target and the hit rate assessment result (probability of hitting the fighting target). The processor compares the hit rate assessment result of the attack action corresponding to each attack orientation with the probability threshold, and identifies those attack orientations that, at the initial position, have a probability greater than or equal to the probability threshold for hitting the fighting target in the initial spatial position. If there are multiple attack orientations that meet the condition (probability greater than or equal to the probability threshold), the processor can further optimize based on factors such as the magnitude of the hit probability, energy consumption, or action continuity, and finally determine one of the attack orientations as the first target orientation. If there is only one attack orientation that meets the probability greater than or equal to the probability threshold, then the above attack orientation can be determined as the first target orientation.

[0064] Optionally, during the adjustment from the initial orientation to the first target orientation, the legged robot can execute a torso orientation adjustment command based on the current initial position. The processor sends control signals to the joint drive module of the legged robot, driving the torso to rotate relative to the lower limbs or the base, so that the forward face of the torso gradually transitions from the current physical orientation to the first target orientation. The above adjustment process aims to ensure that the forward face of the torso (e.g., the area integrating the visual perception module and the main attack limbs) is accurately aligned with the spatial angle indicated by the first target orientation.

[0065] Optionally, during the attack action performed in the direction of the initial spatial position from the first target orientation, when the processor of the legged robot detects that the actual orientation of the forward face of the torso coincides with or is within the allowable error range of the first target orientation, the orientation adjustment is determined to be complete. At this time, the legged robot responds to the first combat command, controlling the target attack limb or the end of the combat limb to move along a predetermined swing trajectory, executing the attack action indicated by the first combat command. Since the first target orientation is a high-probability hit orientation selected based on the initial spatial position, and the attack action is performed in the direction of the initial spatial position, the attack action can accurately target the geometrically bounded surface of the combat target at the moment the command is received, thereby maximizing the probability of hitting the head, torso, or limbs of the combat target.

[0066] In this embodiment, the method described above significantly improves the attack accuracy and success rate of legged robots in combat scenarios by introducing a first target orientation mechanism based on hit probability screening and combining orientation (posture) adjustment and directional execution logic. This mechanism effectively solves the problem of potential directional deviation in the first combat command sent by the user via remote control. Through the legged robot's own intelligent prediction and orientation adjustment, it ensures that the attack action always points towards the effective hit area of ​​the combat target. Furthermore, adjusting the orientation at the initial position rather than performing overall displacement ensures the immediacy of the attack while avoiding additional energy consumption and time delays caused by movement.

[0067] In this embodiment of the application, at least one memory 17 is encoded with instructions that, when executed by at least one processor, cause the legged robot to perform step S108: during the process of adjusting the orientation of the front face of the torso from the initial orientation to the first target orientation, continuously identifying the current spatial position of the fighting object, and continuously filtering for a first updated target orientation in which the probability of the attack action hitting the fighting object in the current spatial position is greater than or equal to a probability threshold; in response to the orientation of the front face of the torso being the first updated target orientation, executing the attack action indicated by the first fighting instruction in the direction of the current spatial position.

[0068] The initial spatial position is used to represent the spatial position of multiple points on the surface of the geometric bounding body of the fighting object in the world coordinate system when the fighting command is received from the remote control device. The current spatial position is used to represent the current position of multiple points on the surface of the geometric bounding body of the fighting object in the world coordinate system. Attack actions include attack actions other than pure area attack actions.

[0069] Optionally, the aforementioned current spatial position can refer to a set of spatial coordinates of multiple points on the geometrically bounding surface of the fighting object, updated in real time in the world coordinate system at any moment during the torso orientation adjustment process of the legged robot. The aforementioned first updated target orientation can refer to the orientation calculated in real time by the processor based on the current spatial position of the fighting object, which satisfies the probability of the attack action hitting a probability threshold greater than or equal to the probability threshold, during the rotation of the torso's forward face from the initial orientation to the first target orientation. The aforementioned first updated target orientation can be dynamically generated and continuously updated iteratively as the current spatial position of the fighting object changes.

[0070] In this embodiment, during the process of adjusting the orientation of the torso's forward face from the initial orientation to the first target orientation, the current spatial position of the combatant can be continuously identified, and a first updated target orientation with a probability greater than or equal to a probability threshold can be continuously filtered to ensure that the attack hits the combatant in the current spatial position. If the orientation of the torso's forward face is the first updated target orientation, the attack action indicated by the first combat command is executed in the direction of the current spatial position.

[0071] Optionally, during the continuous identification of the current spatial position of the fighting object, as the legged robot begins to adjust the forward face of its torso from the initial orientation to the first target orientation, the processor continuously drives the visual perception module to capture visual information at a high frequency. Based on the aforementioned visual information, the processor identifies and tracks the geometric enclosure of the fighting object, and calculates the real-time coordinates of multiple points on the surface of the geometric enclosure in the world coordinate system. These real-time coordinates can be used to construct the current spatial position of the fighting object.

[0072] Optionally, during the continuous screening of the first updated target orientation, the processor utilizes the newly acquired current spatial position of the fighting object, combined with the current torso posture of the legged robot and the kinematic model of the attack action to be executed, to simulate the hit probability of multiple attack orientations in real time. The processor evaluates the degree of overlap between the attack trajectory corresponding to each attack orientation and the current spatial position of the fighting object, screening out attack orientations where the probability of the attack action hitting the fighting object in the current spatial position is greater than or equal to a preset probability threshold. Among the attack orientations that meet the above conditions (probability greater than or equal to the probability threshold), the processor further selects one as the first updated target orientation. This screening process is performed at every moment of torso rotation, ensuring that the selected orientation always adapts to the latest dynamics of the fighting object.

[0073] Optionally, during the attack action performed in the direction of the current spatial position under the first updated target orientation, when the processor detects that the actual physical orientation of the forward face of the legged robot's torso has rotated to coincide with the first updated target orientation (or is within the allowable error tolerance range), the attitude adjustment is determined to be complete. At this time, the processor immediately responds to the first combat command, controls the target attacking limb to move along a predetermined trajectory, and performs the attack action in the direction of the current spatial position of the combatant.

[0074] In this embodiment, the above method significantly improves the attack robustness and hit rate of legged robots when facing moving targets by introducing a dynamic orientation correction mechanism. Traditional static orientation adjustment schemes are prone to failure when the target moves, while this embodiment continuously identifies the current spatial position during rotation and filters the first updated target orientation in real time, thereby achieving high accuracy of attack actions during the movement of the fighting object.

[0075] The legged robot described above in the embodiments of this application will be explained in further detail below.

[0076] As an optional implementation, at least one memory is further encoded with additional instructions, which, when executed by at least one processor, cause the legged robot to: adjust the orientation of the forward face of the torso from an initial orientation to a first target orientation, further comprising: if the first updated target orientation is an orientation in the intermediate process of adjusting from the initial orientation to the first target orientation, then adjusting the orientation of the forward face of the torso from the initial orientation to the first updated target orientation, and in response to the orientation of the forward face of the torso being the first updated target orientation, performing an attack action in the direction of the current spatial position.

[0077] In this embodiment, during the process of adjusting the orientation of the torso's forward face from the initial orientation to the first target orientation, if the first updated target orientation is an orientation that occurs during the intermediate process of adjusting from the initial orientation to the first target orientation, then the orientation of the torso's forward face can be adjusted from the initial orientation to the first updated target orientation. If the orientation of the torso's forward face is the first updated target orientation, an attack action is performed in the direction of the current spatial position.

[0078] Optionally, during the process of determining the first updated target orientation, the processor continuously monitors the real-time generated first updated target orientation while controlling the forward face of the torso to rotate from the initial orientation to the first target orientation. The processor compares the currently calculated first updated target orientation with the preset first target orientation. If the angle value of the first updated target orientation is within the angle range between the initial orientation and the first target orientation, and the angle has not yet reached the final first target orientation, then the first updated target orientation is determined to be an orientation in the intermediate process of adjusting from the initial orientation to the first target orientation.

[0079] Optionally, during the process of adjusting the initial orientation to the first updated target orientation, once the first updated target orientation is determined to be a valid orientation in the intermediate process, the processor modifies the motion control instructions and stops rotating towards the first target orientation. The processor controls the waist joint or torso joint of the legged robot to quickly adjust and lock the actual orientation of the torso's forward face from the current rotation state to the first updated target orientation. The above adjustment process can employ a control algorithm of emergency stop or rapid correction to minimize time delay and ensure that the torso's forward face is precisely aligned with the first updated target orientation in a very short time, preparing the posture for subsequent attack.

[0080] Optionally, during the execution of the attack action under the first updated target orientation, if the processor detects that the actual physical orientation of the torso's forward face has coincided with the first updated target orientation (or is within the allowable error range), it determines that the posture alignment is complete. At this time, the processor no longer waits for the torso to rotate back to the original first target orientation, but directly responds to the first combat command, controls the target attacking limbs to move along a predetermined swing trajectory, and executes the attack action in the direction of the current spatial position of the combatant.

[0081] In this embodiment, the method significantly improves the response speed and attack efficiency of legged robots in dynamic combat scenarios by introducing a dynamic mid-course adjustment mechanism. Related technologies often require a full-angle rotation to execute an attack, which is time-consuming and prone to missing the optimal attack opportunity. However, this embodiment allows for immediate attack execution based on the real-time dynamic spatial position during rotation, once a high-probability mid-course orientation is encountered, greatly reducing the time delay from command reception to action execution. This mechanism not only improves the hit rate against moving targets but also enhances the flexibility and tactical adaptability of legged robots in complex combat environments, achieving a balance between efficiency and accuracy.

[0082] As an optional implementation, at least one memory is further encoded with additional instructions, which, when executed by at least one processor, cause the legged robot to perform an attack action indicated by a first combat command in response to the orientation of the forward face of the torso being a first target orientation, toward the direction of the current spatial position. This includes: if, during the process of adjusting the orientation of the forward face of the torso to the first target orientation, the probability of the attack action hitting the combat target at the current spatial position is less than a probability threshold, and the first updated target orientation is the orientation after the adjustment from the initial orientation to the first target orientation, then the orientation of the forward face of the torso is adjusted from the initial orientation to the first updated target orientation, and in response to the orientation of the forward face of the torso being the first updated target orientation, the attack action is performed in the direction of the current spatial position.

[0083] In this implementation, during the execution of the attack action instructed by the first combat command in the direction of the current spatial position under the first target orientation, if the probability of the attack hitting the combat target at the current spatial position is less than a probability threshold during the adjustment of the torso's forward orientation to the first target orientation, and the first updated target orientation is the orientation after the initial orientation is adjusted to the first target orientation, then the torso's forward orientation can be adjusted from the initial orientation to the first updated target orientation. If the torso's forward orientation is the first updated target orientation, then the attack action can be executed in the direction of the current spatial position.

[0084] Optionally, during the process of determining that the probability of each attack orientation is less than a probability threshold during the adjustment of the orientation to the first target orientation, the processor continuously acquires the current spatial position of the combatant while controlling the rotation of the torso's forward face from the initial orientation to the first target orientation, and calculates the hit probability of attack actions under various candidate orientations in real time. The processor monitors the probability of an attack action corresponding to a specific attack orientation hitting the combatant in the current spatial position when the torso's forward face approaches or reaches the preset first target orientation. If the calculation result shows that the hit probability is less than the preset probability threshold at the moment the torso's forward face is adjusted to the first target orientation, it is determined that executing an attack at the first target orientation cannot guarantee an effective hit on the target, thereby triggering the subsequent overshoot correction process.

[0085] Optionally, in determining the first updated target orientation as the orientation following the first target orientation, the processor simulates the hit probability for subsequent angles beyond the first target orientation based on the continuously changing current spatial position of the fighting object and the continued rotational movement trend of the legged robot. The processor searches for angle points that exist after the first target orientation (i.e., within the angle range where the rotation angle is greater than the first target orientation) and make the hit probability of the attack action greater than or equal to a probability threshold. If such an angle point exists and is located in the rotational forward direction of the first target orientation, then that angle point is determined as the first updated target orientation.

[0086] Optionally, during the process of adjusting the orientation of the torso's forward face to the first updated target orientation, once the first updated target orientation is determined to be a valid orientation after the first target orientation, the processor adjusts the motion control strategy, instructing the legged robot not to stop or perform an attack at the first target orientation, but to maintain rotational inertia or actively accelerate, continuing to rotate the torso's forward face further from the first target orientation to the first updated target orientation. The processor monitors the actual angle of the torso's forward face in real time until the actual angle coincides with the first updated target orientation (or is within the allowable error range), completing the final attitude lock.

[0087] Optionally, during the execution of the attack action under the first updated target orientation, when the processor determines that the actual physical orientation of the torso's forward face has been precisely aligned with the first updated target orientation, the attitude adjustment is considered complete. At this time, the processor immediately responds to the first combat command, controlling the target attacking limbs to move along a predetermined trajectory and execute the attack action in the direction of the current spatial position of the combatant.

[0088] In this embodiment, the method described above significantly improves the hit rate of legged robots when facing moving targets by introducing an overshoot correction mechanism based on dynamic assessment of hit probability. Related technologies often mechanically execute an attack upon reaching a preset first target orientation, which can easily lead to missed attacks due to target displacement. However, this embodiment allows the legged robot to "rotate a little further" during rotation until it finds a truly high-probability hit position (the first updated target orientation) before launching an attack. This method overcomes the limitations of static preset orientations in dynamic environments, ensuring that the attack action always targets the most threatening or most vulnerable attack orientation, greatly enhancing the legged robot's intelligent decision-making ability and tactical execution accuracy in complex combat scenarios.

[0089] As an optional implementation, at least one memory is further encoded with additional instructions, which, when executed by at least one processor, cause the legged robot to perform an attack action indicated by a first combat command in response to the orientation of the forward face of the torso being a first target orientation, in the direction of the current spatial position. This includes: if, during the adjustment of the orientation of the forward face of the torso to the first target orientation, the probability of the attack action hitting the combat object at the current spatial position is less than a probability threshold, and no probability of the attack action hitting the combat object at the current spatial position being greater than or equal to a probability threshold is identified in other orientations, then after the orientation of the forward face of the torso is adjusted to the first target orientation, the attack action is performed in the direction of the current spatial position.

[0090] In this embodiment, when the forward face of the torso is facing the first target orientation, and the attack action is executed in the direction of the current spatial position as indicated by the first combat command, if the probability of the attack action hitting the combat target at the current spatial position is greater than or equal to a probability threshold during the process of adjusting the forward face of the torso to the first target orientation, then the attack action can be executed in the direction of the current spatial position after the forward face of the torso is facing the first target orientation.

[0091] Optionally, during the entire process of controlling the torso's forward face to rotate from the initial orientation to the first target orientation, the processor continuously acquires the current spatial position of the combatant in real time, and evaluates the hit rate of various possible attack orientations, both current and future, based on the current spatial position. If, throughout the entire rotation interval from the initial orientation to the first target orientation, the probability of the attack action corresponding to the evaluated attack orientation hitting the combatant at the current spatial position is less than a preset probability threshold, it is determined that under the current dynamic situation, no posture on the preset path can provide a suitable striking effect.

[0092] Optionally, in determining that no other orientation exists that makes the probability greater than or equal to the probability threshold, after determining that the probability of each attack orientation from the initial orientation to the first target orientation is less than the probability threshold, the processor can expand the search range to perform a global hit probability assessment on each potential attack orientation within a 360-degree full-angle range (or within the effective rotation range allowed by the mechanical structure) of the torso's forward face. The processor performs spatial overlap calculations between the current spatial position of the fighting object and the attack trajectory under each potential orientation. If, among all the global candidate attack orientations, there is no orientation that makes the probability of the attack hitting the fighting object at the current spatial position greater than or equal to the probability threshold, it is determined that at the current moment, there is no first updated target orientation that satisfies the high hit rate condition, thereby triggering the backup execution logic.

[0093] Optionally, during the process of adjusting the orientation of the torso's forward face to the first target orientation, even if the global search does not find an orientation that satisfies a high hit probability, the processor still follows the preset tactical intent of the instruction and continues to control the torso joints of the legged robot to completely adjust the forward face of the torso from the current rotation state to the first target orientation, ensuring that the legged robot completes the predetermined action preparation posture, and does not interrupt the action process or change the preset action execution benchmark due to a low hit probability, thus maintaining the continuity and standardization of action execution.

[0094] Optionally, during the attack maneuver performed in the direction of the current spatial position from the first target orientation, when the processor detects that the actual physical orientation of the torso's forward face has aligned with the first target orientation, the attitude adjustment is considered complete. At this point, the processor immediately responds to the first combat command, without waiting for or searching for other better opportunities, and directly controls the target attacking limb to move along a predetermined swing trajectory, performing the attack maneuver in the direction of the current spatial position of the combatant. Although the probability of the attack maneuver executed by the above method hitting the combatant is less than the probability threshold, the legged robot maintains its combat rhythm by executing the above attack maneuver, which is an available tactical action under the current dynamic constraints.

[0095] In this embodiment, the method significantly improves the tactical robustness of legged robots in extreme dynamic environments by introducing a fallback execution mechanism after a global search failure. Related technologies often abandon attacks or become stuck in decision-making stagnation due to the inability to find a high-probability attack direction. However, this embodiment clearly defines the action guidelines when no attack direction with a probability greater than or equal to a probability threshold exists, namely, forcibly executing an attack towards a preset first target direction. This not only prevents the legged robot from missing opportunities due to over-calculation or perfectionism but also ensures the complete execution of the first combat command, maintaining the continuity of combat.

[0096] As an optional implementation, at least one memory is further encoded with additional instructions, which, when executed by at least one processor, enable the legged robot to acquire the initial spatial position of a fighting object in the world coordinate system from visual information by performing the following steps: in response to a received first fighting instruction including information about the fighting object, selecting a fighting object that satisfies the information about the fighting object from at least one fighting object in the acquired visual information, and acquiring the initial spatial position of the fighting object that satisfies the information about the fighting object from the visual information; receiving an instruction to determine the fighting object before receiving the first fighting instruction, and acquiring the initial spatial position of the fighting object indicated by the instruction to determine the fighting object from the acquired visual information; in response to a received first fighting instruction including fighting direction information, determining the fighting object from the acquired visual information according to the fighting direction pointed to by the fighting direction information, and acquiring the initial spatial position of the fighting object in the fighting direction pointed to by the fighting direction information.

[0097] The information about the combat target mentioned above can refer to the identity data or identification code directly included in the first combat command sent by the remote control device, used to identify or specify a specific combat target. This information can correspond to a pre-bound target identifier. For example, in a combat scenario where a legged robot is fighting two combat targets (e.g., combat target A and combat target B), the remote control device's display interface can show two selectable target icons, labeled "Target 1" and "Target 2," respectively. "Target 1" corresponds to the target identifier of robot A, and "Target 2" corresponds to the target identifier of robot B. When the user selects to attack "Target 1" and sends the first combat command, the "combat target information" carried in the first combat command is the target identifier of robot B. After parsing the first combat command, the legged robot can filter out the combat target with the target identifier of robot B from the visual information and use this combat target as the combat target to execute the attack action indicated by the first combat command.

[0098] The aforementioned command to identify the fighting target can refer to an independent control signal sent separately by the remote control device before the first fighting command is sent, used to lock onto or identify the target (fighting target) to be attacked in the current attack action. For example, in a fighting scenario where a legged robot is fighting two fighting targets (e.g., fighting target A and fighting target B), the user can align the crosshairs on the display screen with one of the two fighting targets and lock onto it using the "lock button" on the remote control device or the "click to confirm" operation on the screen. The remote control device sends the command to identify the fighting target to the legged robot. After receiving the command to identify the fighting target, the legged robot immediately locks onto the fighting target indicated by the command and obtains the current initial spatial position of the fighting target. Subsequently, the user sends the first fighting command, and the legged robot can directly execute the attack action indicated by the first fighting command on the locked fighting target without specifying the target again in the attack command.

[0099] The aforementioned combat direction information refers to the angular data or relative direction vector included in the first combat command, indicating the attack direction, rather than a specific target identifier. This combat direction information can be used to instruct the legged robot to find the nearest or most prominent combat target in a specific direction as the target for the attack. For example, if a user presses the "right hook" button on a remote control, the associated attack command includes "combat direction information," such as a direction vector indicating "30 degrees to the right in front." After parsing this direction vector, the legged robot searches for combat targets located within the cone-shaped area of ​​"30 degrees to the right in front" in the scene acquired by the visual perception module. If multiple combat targets are detected in this cone-shaped area, the legged robot can select the closest one, or select one according to a preset strategy, and use the initial spatial position of the selected combat target as the location of the target to be attacked. This method is suitable for scenarios where the user only needs to focus on striking in a certain direction without needing to precisely specify which combat target.

[0100] In this implementation, if the received first fighting command includes information about the fighting object, a fighting object satisfying the aforementioned information can be selected from at least one fighting object in the acquired visual information, and the initial spatial position of the fighting object can be obtained from the visual information. Before receiving the first fighting command, a command from the fighting object can be received, and the initial spatial position of the fighting object indicated by the command can be obtained from the acquired visual information. If the received first fighting command includes fighting direction information, the fighting object can be determined from the acquired visual information based on the fighting direction indicated by the fighting direction information, and the initial spatial information of the fighting object in the indicated fighting direction can be obtained.

[0101] Optionally, when the received first combat command contains information about the combat object, the legged robot executes selection logic based on identifier matching. The processor can identify and extract each visible combat object and its corresponding identity from the visual information obtained from the visual perception module. The processor compares the combat object information contained in the first combat command with the identity identifiers of each visible combat object, and selects the combat object with matching identity information as the combat object to be attacked. Once the target combat object is determined, the processor parses the three-dimensional geometric features of the combat object from the visual information, calculates the position coordinates of points on the surface of the geometric bounding body of the combat object in the world coordinate system, and thus obtains the initial spatial position of the combat object that meets the information of the combat object. The above mode is applicable to scenarios where one or more combat objects exist, and target designation is achieved through precise identity binding.

[0102] Optionally, before receiving the first combat command, if the legged robot receives a command to identify a combat target, it executes pre-lock selection logic. This command to identify the combat target can be triggered by a lock-on operation issued by the user via a remote control device, used to determine the target of the attack action among multiple combat targets. After receiving the command to identify the combat target, the processor identifies the specific combat target indicated by the command from visual information and marks it as the current lock target. Based on the visual information of the current lock target, the processor calculates its spatial coordinates in the world coordinate system as the initial spatial position of the identified target. This mode is suitable for complex scenarios with multiple combat targets, ensuring the uniqueness and accuracy of the combat target for the attack action through a two-step operation of locking before attacking.

[0103] Optionally, when the received first combat command contains combat direction information but not a specific lock-on identifier, the legged robot executes selection logic based on a direction range. The processor parses the combat direction information in the first combat command to determine the pointing combat direction. This combat direction can be represented as an angular range or a vector direction. The processor searches for combat objects within the combat direction region pointed to by the aforementioned combat direction in the visual information. Based on a preset strategy (e.g., the nearest distance principle or priority principle), the processor determines a combat object from the combat objects within the aforementioned combat direction region to target with the attack action. The processor calculates the positions of multiple points on the geometric bounding surface of the aforementioned combat object in the world coordinate system as the initial spatial position of the combat object. This mode is suitable for scenarios where the user only needs to specify the general direction of the attack action rather than a specific individual, improving operational convenience.

[0104] In the embodiments of this application, the target selection and position acquisition mechanisms in the above three scenarios significantly improve the target recognition capability and operational flexibility of legged robots in the environment. By supporting direct identifier matching, pre-locking, and direction range selection, legged robots can adapt to various scenarios from single-target combat to multi-target melee, meeting the operating habits and needs of different user groups.

[0105] As an optional implementation, at least one memory further encodes additional instructions, which, when executed by at least one memory, cause the legged robot to: after pre-calculating a hit rate assessment of multiple attack orientations, further include: if a first target orientation is not selected from the multiple attack orientations, and the probability of the attack action hitting the fighting object at the initial spatial position is greater than or equal to a probability threshold, then stop receiving movement commands triggered by the remote control device, or stop executing received movement commands triggered by the remote control device and move from the initial position to the first target position; and at the first target position, perform a hit rate assessment of multiple attack orientations through pre-calculation; if a first target orientation is selected from the multiple attack orientations, and the probability of the attack action hitting the fighting object at the initial spatial position is greater than or equal to a probability threshold, then at the first target position, adjust the orientation of the torso's forward face from the initial orientation to the first target orientation, and in response to the orientation of the torso's forward face being the first target orientation, perform an attack action in the direction of the initial spatial position.

[0106] In this embodiment, if no first target orientation is selected from multiple attack orientations that has a probability greater than or equal to a probability threshold for hitting the fighting object in the initial spatial position, then receiving movement commands triggered by the remote control device can be stopped, or executing received movement commands triggered by the remote control device can be stopped, and the robot can move from the initial position to the first target position. At the first target position, the hit rate of multiple attack orientations can be evaluated through calculation. If a first target orientation is selected from multiple attack orientations that has a probability greater than or equal to a probability threshold for hitting the fighting object in the initial spatial position at the first target position of the legged robot, then the orientation of the torso's forward face can be adjusted to the first target orientation at the first target position. This allows the attack action to be performed in the direction of the initial spatial position.

[0107] Optionally, the first target position can be the optimized spatial coordinates to which the legged robot actively moves to achieve an effective hit when the hit rate assessment result does not reach the probability threshold. This first target position can be calculated using a spatial search algorithm based on the maximum effective range of the attack action, the initial position of the fighting object, and the terrain constraints of the environment. This ensures that when the attack action is performed at this first target position, the geometric intersection between the swing trajectory and the fighting object satisfies the probability threshold. It should be noted that in this embodiment, the pre-calculation process is not repeated at the first target position. Instead, the attack is performed with the first target facing the target to shorten the response delay and improve the immediacy and determinism of the user-issued fighting commands.

[0108] Optionally, the first target orientation can satisfy the following three conditions: the first condition is determined in response to the first combat command; the second condition is that the probability of the attack action hitting the combat target is greater than or equal to a probability threshold after pre-calculation; and the third condition is the orientation of the forward face of the torso corresponding to the initial action of the attack action.

[0109] Optionally, regarding the first condition (determined in response to the first combat command), the analysis is as follows: The first target orientation can be directly generated by the first combat command received by the legged robot, which serves as the trigger source for the attack action. After parsing the first combat command, the corresponding swing trajectory model and associated parameters can be retrieved from the action library based on the command content, thereby limiting the candidate range of the first target orientation. For example, when the first combat command is a right straight punch, the orientation of the forward face matching the swing trajectory of the right straight punch can be considered, while the orientation logic of other attack actions (such as a hook punch) is not allowed to be introduced. The first condition ensures that the first target orientation is aligned with the user's command intent of the first combat command, avoiding unauthorized changes to the attack intent and ensuring the predictability and consistency of the control intent.

[0110] Optionally, regarding the second condition mentioned above (that the probability of the attack hitting the target is greater than or equal to a probability threshold after pre-calculation), the analysis is as follows: The first target orientation is not arbitrary, but rather a specific orientation whose hit probability reaches or exceeds the combat threshold after pre-calculating the trajectory of the end of the fighting limb and the position of the fighting target. The pre-calculation process is based on the initial position and initial spatial position, calculating the intersection probability of the swing trajectory and the geometric enclosure of the fighting target under each candidate attack orientation at that position. Finally, the candidate attack orientation with the highest probability, greater than or equal to the aforementioned probability threshold, is selected. For example, if the swing trajectory penetrates the opponent's head under an orientation with a 25° right rotation relative to the torso, the hit probability is 0.96; while under orientations with a 15° or 35° right rotation relative to the torso, the probability is less than 0.65, then the final selected candidate attack orientation is a 25° right rotation relative to the torso. In this embodiment of the application, the second condition mentioned above can ensure that the attack orientation of the selected candidate has actual tactical effectiveness, rather than just being based on directional alignment, thus avoiding misjudgments of "seemingly aligned but actually missing the target" and improving the attack success rate and the decision-making reliability of the legged robot.

[0111] Optionally, regarding the third condition mentioned above (the orientation of the forward face of the torso corresponding to the initial movement of the attack), the analysis is as follows: The first target orientation can spatially correspond to the initial movement of the attack. That is, the first target orientation can be the orientation of the forward face required by the starting point of the swing trajectory, rather than the orientation at any moment during the attack (e.g., at the moment of striking or in the recovery posture). The initial movement is the first identifiable preparatory posture in the attack sequence. For example, the forward lean of the shoulder in a straight punch, the backward shift of the center of gravity and torso twist in a sweeping kick. The orientation of the forward face when executing the initial movement determines the initial direction of motion at the end of the attack, ensuring that the legged robot is in the correct pre-swing posture before executing the attack, so that the subsequent trajectory tracking has physical consistency. For example, if the initial posture of the sweeping kick requires the torso to twist 45° to the left, then regardless of where the fighting opponent is on the right, the first target orientation must be 45°, rather than a straight line towards the fighting opponent. In the embodiments of this application, the third condition mentioned above can effectively prevent the legged robot from confusing "facing the fighting object" with "facing at the start of the action", ensuring the executability of the attack action in the mechanical structure and the standardization of the action library.

[0112] In this embodiment, the complete definition of the first target orientation, constituted by the above three conditions, is a physical starting posture that can be triggered by user commands, has been pre-calculated and verified to have hit capability, and strictly matches the attack action. These three conditions together ensure that the legged robot, before executing the attack, not only conforms to the operational intent and possesses technical feasibility, but also maintains the standardization and consistency of the action design.

[0113] The hit rate assessment results obtained through pre-calculation can include two cases: one is the hit probability assessment based on traversing multiple attack orientations for candidates, and the other is the static feasibility judgment based on the inherent effective distance of the action.

[0114] Regarding the first scenario, the hit rate assessment result can be the probability of hitting the target in the initial spatial position under each candidate attack orientation, assuming the torso remains in its initial position. This probability is calculated by pre-calculating the trajectory of the attack action indicated by the first combat command within a pool of candidate attack orientations. The pre-calculation process can be based on a preset end-point trajectory model of the attack action, combined with the geometric bounding volume of the target's initial spatial position. Frame-by-frame calculations are performed to determine the spatial intersection between the swing trajectory and the target's body under different candidate attack orientations, including intersection volume, penetration depth of key areas, and trajectory endpoint coverage effectiveness. Finally, the attack orientation with the highest hit probability among the candidate orientations is selected as the first target orientation. In summary, the hit rate assessment result in the first scenario can be used to determine whether an effective hit can be achieved by adjusting the orientation of the torso's forward face from the initial position.

[0115] Regarding the second scenario mentioned above, the hit rate assessment result can be a feasibility judgment based on the spatial distance between initial spatial positions, combined with the preset maximum effective radius or trajectory endpoint range of the attack action corresponding to the first combat command, without orientation traversal. The preset maximum effective radius can be the farthest distance boundary that the end-effectors (e.g., fist, foot, or weapon tip) of the legged robot can reach under physical structural constraints. This radius can be fixed by the joint travel, limb length, and motion library parameters of the legged robot, defining the spherical range that the attack action can cover without orientation adjustment. The trajectory endpoint range can be a geometric region consisting of the theoretical final positions of the end-effectors when the attack action defined by the first combat command is completed. This geometric region can be a point, line segment, arc, or sector, determined by the endpoint path of the attack action's trajectory. It is used to determine whether the combat target falls within the effective contact area upon completion of the attack, without relying on intermediate trajectory shapes, but rather using final state coverage as the judgment criterion. The above hit rate assessment does not take into account the specific orientation, but rather determines whether the initial spatial position is within the geometrically accessible physical boundaries of the attack action.

[0116] In this embodiment, if the legged robot receives the first combat command, it can initiate an attack feasibility pre-calculation process based on spatial relative relationships. Using the initial spatial position and initial position as input, it uses two optional pre-calculation mechanisms (combined with orientation traversal or no orientation traversal) to determine whether the attack action has the potential to hit, and decides whether to directly adjust the orientation of the forward face or to move first and then execute, thereby improving the determinism of the combat command response and the smoothness of the user's operation while ensuring the effectiveness of the attack.

[0117] Optionally, during the pre-calculation of the attack action, two optional pre-calculation mechanisms can be used depending on the type of the first combat command. The first optional pre-calculation mechanism is a static distance pre-calculation without orientation traversal. During the pre-calculation process, the straight-line distance between the initial spatial positions can be compared with the preset maximum effective radius or trajectory endpoint range of the attack action. If the initial spatial position is within the aforementioned maximum effective radius or trajectory endpoint range, it is determined to be hit; otherwise, it is considered hitless. For example, if the attack action is a straight punch with a preset maximum effective radius of 1.2 meters, and the distance to the center of the combatant's torso is 1.0 meter, it is determined to be hit; if the distance is 1.4 meters, it is determined to be hitless. If the attack action is a sweeping kick, the trajectory endpoint range of the sweeping kick is a fan-shaped area with a radius of 1.3 meters centered on the foot. If the combatant's leg is within this fan-shaped area, even if the torso is not directly facing it, it can be considered hitable. In the embodiments of this application, the above-mentioned static distance pre-calculation mechanism without orientation traversal has high computational efficiency and is suitable for first combat commands with fixed swing trajectories and clear range of action of attack actions. It can effectively reduce the latency of legged robots and improve the user's "point-and-response" operation intuition.

[0118] The second optional pre-calculation mechanism is a multi-candidate attack orientation hit pre-calculation with orientation traversal. Starting from the initial position as the origin, and using a preset degree (e.g., 1° or 5°) as the step size, it traverses each candidate attack orientation in a library of attack orientations (e.g., 0° to 360°). Under each candidate attack orientation, it generates the end trajectory of the fighting limb when executing the attack action and performs collision detection with the geometric bounding box of the fighting object. The hit determination is based on the tactical value classification of the contact area: if the end trajectory penetrates the geometric bounding box of the fighting object's head or core torso (e.g., chest cavity or abdomen), it is considered a "high-value hit" and assigned a probability weight of 0.9; if the end trajectory penetrates the fighting object's shoulder, thigh, arm, hand, or other limbs or their extremities, it is considered a "low-value hit" and assigned a probability weight of 0.4; if the end trajectory only grazes the edge of the fighting object's outer shell or does not make contact, the probability is 0. The probability of hitting various parts of the combat target from each candidate attack direction can be weighted and averaged to determine the hit rate assessment result for that candidate attack direction. The attack direction of the candidate with the highest probability threshold, where the hit rate assessment result is greater than or equal to a probability threshold, can be used as the first target direction. In the embodiments of this application, the above pre-calculation mechanism is suitable for high-precision directional attacks with complex swing trajectories, ensuring that the torso can find a suitable first target direction when performing the attack action in the initial position, thereby enhancing the accuracy and tactical adaptability of the attack.

[0119] Optionally, if the hit rate assessment result is that the probability is greater than or equal to the probability threshold, orientation adjustment is performed. If a pre-calculation mechanism with orientation traversal is used, the first target orientation is the specific angle that produces the highest hit probability among the various pre-calculated candidate attack orientations; if a static pre-calculation mechanism without orientation traversal is used, the first target orientation can be a preset default alignment orientation. That is, the horizontal projection direction from the central axis of the legged robot's torso to the initial spatial position is directly calculated from the angle between the two position vectors in the horizontal plane. For example, the first target orientation under the above static pre-calculation mechanism without orientation traversal can be a preset orientation in which the legged robot's torso needs to face the fighting object when performing the attack action. This design ensures that the attack action is aligned with the target before it is initiated. It avoids "missing" the attack action due to incorrect orientation, increases the user's confidence in the accuracy of the attack, and enhances the immersive control experience.

[0120] Optionally, if the hit rate assessment result is that the probability is less than the probability threshold, the system can immediately stop receiving subsequent movement commands sent by the remote control device, or suspend received but not yet executed movement actions. This stopping operation aims to prevent the user from continuously inputting conflicting commands while the legged robot is about to execute an attack action, avoiding erratic movements or wasted energy due to hesitation. For example, if the user clicks "move forward" and "move left" consecutively after the first combat command is issued, without the above stopping operation, the legged robot may be interfered with before adjusting its orientation, leading to attack delays or failures. Through this stopping operation, the principle of "pre-calculation priority" can be established, ensuring the integrity of attack decisions, allowing the user to perceive that the legged robot is "carefully preparing" rather than blindly executing, thus enhancing the sense of ritual and controllable expectation of operation.

[0121] Optionally, if the hit rate assessment result is that the probability is less than the probability threshold, the search for the first target location is initiated. The first target location can be found within a circle centered on the initial spatial position and with the maximum effective radius of the attack action as its radius. This is achieved by using a spatial algorithm to find a spatial point that is closest to the initial spatial position and whose intersection probability with the geometric enclosure of the fighting limb's end-effector is greater than or equal to the probability threshold. Alternatively, the first target location can be a pre-set location. For example, moving a preset distance towards the initial spatial position of the fighting object can shorten the distance between the legged robot and the fighting object, increasing the probability of the attack hitting the object. For instance, if the current distance between the fighting object and the legged robot is 1.5 meters, and the attack action is a spinning kick with a maximum effective radius of 1.3 meters, a spatial point 0.6 meters in front of the initial position of the torso can be calculated. This reduces the distance between the legged robot and the fighting object to 0.9 meters, at which point the end-effector's trajectory can reliably penetrate the geometric enclosure of the fighting object, increasing the hit probability. This spatial point can then be designated as the first target location. The legged robot can move from its initial position to the first target position mentioned above along the shortest path.

[0122] Optionally, after reaching the first target position, the legged robot does not need to perform a new round of pre-calculation and can directly adopt the first target orientation when the probability is greater than or equal to the probability threshold. That is, if the candidate attack orientations were traversed during the pre-calculation process, the orientation with the highest probability of hitting the fighting object among the multiple candidate attack orientations is used as the first target orientation; if the candidate attack orientations were not traversed during the pre-calculation process, but rather a static position judgment was performed, the default alignment orientation can be used. In the embodiments of this application, the above process avoids secondary pre-calculation because the hit probability has been ensured to be greater than the generalization threshold through geometric constraints during the position adjustment stage. Repeated calculation is not only useless but also introduces unnecessary delays. The above method significantly shortens the response cycle of fighting commands, allowing the attack action to be executed immediately after the movement is completed. The user does not perceive the delay after issuing the command, enhancing the continuity and entertainment value of the fighting rhythm.

[0123] As an optional implementation, the fighting target includes a head and a torso. At least one memory is also encoded with additional instructions, which, when executed by at least one processor, cause the legged robot to: pre-calculate and evaluate the hit rate of multiple attack orientations, including: if the execution of the attack action can hit the head or torso of the fighting target, determining that the probability of the attack action hitting the fighting target is greater than or equal to a probability threshold; if the execution of the attack action does not hit the head or torso of the fighting target, determining that the probability of the attack action hitting the fighting target is less than a probability threshold.

[0124] In this embodiment, if the attack action can hit the head or torso of the fighting target, it indicates that the probability of the attack hitting the fighting target is greater than or equal to a probability threshold. If the attack action fails to hit the head or torso of the fighting target, it indicates that the probability of the attack hitting the fighting target is less than the probability threshold. This embodiment proposes a judgment logic for a legged robot to determine the relationship between the probability of an attack hitting the fighting target and a preset probability threshold. The processor quantifies the probability of a hit by pre-calculating the spatial overlap between the target attack limb (or the end of the fighting limb) and the key parts of the fighting target (head, torso). If spatial overlap occurs, the probability of a hit is determined to be greater than or equal to the probability threshold; if there is no spatial overlap, the probability of a hit is determined to be less than the probability threshold. This judgment logic applies not only to static pre-calculation based on the initial position (first pre-calculation case) but also to dynamic pre-calculation based on the real-time dynamic position (second pre-calculation case), thereby ensuring that the robot can make accurate attack decisions in different time dimensions.

[0125] Optionally, during the determination process, the processor focuses on whether the attack action can effectively cover the high-value hit areas (critical parts, such as the head and torso) of the fighting object. Whether it is a first pre-calculation based on the initial spatial position and the initial position (static initial state), or a second pre-calculation based on the real-time current spatial position and the initial position of the fighting object (dynamic real-time state), as long as the first hit rate evaluation result obtained from the first pre-calculation or the second hit rate evaluation result obtained from the second pre-calculation shows that executing the attack action can hit any one or more of the fighting object's head or torso, the processor determines that the probability of the attack action hitting the fighting object is greater than or equal to the probability threshold.

[0126] Optionally, for the first pre-calculation scenario, if the pre-calculated attack trajectory intersects spatially with the head or torso of the combatant at the initial spatial position while the legged robot remains stationary, it is considered a high-probability hit. For the second pre-calculation scenario, if the current spatial position of the combatant is acquired in real time during the legged robot's adjustment of orientation or movement, and the pre-calculated attack trajectory intersects spatially with the head or torso of the combatant at that current spatial position, it is also considered a high-probability hit. The above criteria are based on the fact that the head and torso are the main and stable hit areas of the combatant, and hitting these hit areas means that the attack action has substantial tactical value.

[0127] Optionally, if the hit rate assessment shows that the attack action fails to cover the target area (e.g., head or torso) of the fighting object, it is determined to be a low-probability hit. In the first or second pre-calculation scenario, if the processor verifies that even if the legged robot fully executes the attack action, the swing trajectory of the target attacking limb or the end of the fighting limb cannot have any spatial overlap with the head and torso of the fighting object (i.e., the attack trajectory is completely deviated from the head and torso), the processor determines that the probability of the attack action hitting the fighting object is less than the probability threshold.

[0128] Optionally, for the first pre-calculated scenario, if the attack trajectory does not reach the head or torso of the fighting target from the initial position, it indicates that the initial position is not suitable for initiating the attack action, and the robot needs to move before executing the attack action or execute the attack action directly. For the second pre-calculated scenario, during dynamic tracking, if the swing trajectory does not reach the head or torso of the fighting target in real time, it indicates that the fighting target has moved outside the attack range corresponding to the swing trajectory or is in a defensive posture, and the probability of hitting the target when executing the attack action is low. The above judgment logic ensures that the legged robot can promptly identify the inefficiency of the attack when the head and torso of the fighting target are not hit.

[0129] In this embodiment, the method described above provides a concise and efficient hit probability assessment method for legged robots by introducing a spatial overlap determination mechanism based on the head and the target's torso. This mechanism simplifies the complex physical collision probability into a clear geometric existence judgment (hitting or not hitting key parts), which not only reduces computational complexity and improves processor response speed, but also ensures the tactical rationality of attack decisions. By being compatible with the first pre-calculation (initial state) and the second pre-calculation (real-time state), it can make predictions before the attack is initiated and make real-time corrections during the attack execution, significantly improving the attack accuracy and reaction efficiency of legged robots in dynamic combat scenarios.

[0130] Optionally, the process of determining the probability of hitting the fighting object in the hit rate assessment results (e.g., the first hit rate assessment result and the second hit rate assessment result) can be divided into the following three determination scenarios based on the intersection relationship between the swing trajectory and the spatial structure of the fighting object: the first scenario is that the swing trajectory passes through the outer shell of the fighting object; the second scenario is that the swing trajectory contacts the surface of the outer shell of the fighting object but does not pass through the outer shell; and the third scenario is that the swing trajectory does not intersect with the outer shell of the fighting object (i.e., it does not contact the outer shell).

[0131] Optionally, in the first scenario described above, the swing trajectory of the target limb not only intersects the outer shell of the fighting object in spatial geometry, but at least a portion of the swing trajectory lies within the enclosed volume of the fighting object formed by its outer shell. This means that the attack action possesses direct physical penetration or substantial impact conditions, and is determined to be a high-probability (probability of hitting the fighting object greater than or equal to a probability threshold) effective hit. The above scenario corresponds to the swing trajectory directly pointing towards the core area (e.g., head area, torso core) or major force-bearing area (joints) of the fighting object.

[0132] For example, the first scenario described above could include head strikes, trunk strikes, and joint damage. For instance, regarding a head strike, the trajectory of a legged robot's punch passes through the surface of the target's head shell and extends into the internal space of the target's skull model, thus being considered a strike to the core of the head. Regarding a trunk strike, the trajectory of a robot's kick passes through the target's abdominal shell and enters the internal volume of the target's torso column, thus being considered a strike to the core of the torso. Regarding joint damage, the trajectory of a robot's elbow strike passes through the target's shoulder shell and enters the internal volume of the shoulder joint connection, thus being considered a strike to the joint pivot.

[0133] Figure 3 This is a schematic diagram illustrating the hit probability of a legged robot according to an embodiment of this application, as shown below. Figure 3As shown, after pre-calculation, it is determined that the swing trajectory of the legged robot's left straight punch will pass through the surface of the fighting object's head shell, that is, directly hit the fighting object's head. This indicates that the probability of the current left straight punch hitting the fighting object is greater than or equal to the probability threshold.

[0134] Optionally, in the second scenario described above, the swing trajectory of the target fighting limb intersects with the outer surface of the fighting object, but this intersection is limited to contact with geometric surfaces or lines, and the trajectory line segment does not intrude into the internal volume of the fighting object. This scenario can manifest as grazing, edge striking, or contact only with non-core limb extremities. In these cases, although the legged robot is in contact with the fighting object, the energy transfer efficiency is low or the striking area is a non-core region, resulting in a medium or low probability of a successful hit. The success rate depends on the importance of the contact area.

[0135] For example, the second scenario mentioned above can include edge grazing, extremity contact, lateral sweeping, and limb strikes. Specifically, for edge grazing, the swing trajectory of the legged robot's punch only grazes the edge outline of the target's head shell, without penetrating the internal volume of the skull, and is considered as only contacting the edge of the shell. For extremity contact, the swing trajectory of the legged robot's kick only touches the surface of the target's outstretched arm, without reaching the target's torso or head, and is considered as contacting the opponent's arm. For lateral sweeping, the swing trajectory of the legged robot's punch only sweeps across the lateral boundary of the target's chest shell, without penetrating the chest cavity, and is considered as only tangential to the surface of the shell. For limb strikes, the legged robot's kick only hits the lower edge of the target's calf, without reaching the thigh or core of the torso, and is considered as only contacting the outer shell of the opponent's limb.

[0136] Figure 4 This is a schematic diagram illustrating the hit probability of another legged robot according to an embodiment of this application, such as... Figure 4 As shown, after pre-calculation, it was determined that the swing trajectory of the legged robot's punch only touched the opponent's arm, without contacting the opponent's core area. This indicates that the legged robot only touched the opponent's outer shell. In this embodiment, since the legged robot only touched the opponent's limbs and not the torso or head, it can be concluded that the probability of hitting the opponent is less than the probability threshold.

[0137] Optionally, in the third scenario described above, the swing trajectory of the target's fighting limbs lies entirely outside the three-dimensional model of the fighting object in space, with no geometric point, line, or surface contact between them. This means the attack completely misses its target and is judged as a zero-probability effective hit. This scenario can occur when the distance is too great, the orientation deviation is too large, or the target's movement causes trajectory misalignment.

[0138] For example, the third scenario mentioned above could include: significant misses, deviations due to insufficient or excessive distance, and complete misalignment. Specifically, in the case of a significant miss, the legged robot's punching trajectory passes above or to the side of the opponent's head, without contacting the surface of the opponent's head, torso, or limbs. In the case of deviations due to insufficient or excessive distance, the legged robot's kicking trajectory passes under the opponent's feet, without touching any of the opponent's limbs or torso. In the case of complete misalignment, when the legged robot performs a sweeping leg motion, the leg's trajectory is completely off-center from the area where the opponent is standing, with no intersection with the opponent's body shell.

[0139] Figure 5 This is a schematic diagram illustrating the hit probability of another legged robot according to an embodiment of this application, such as... Figure 5 As shown, after pre-calculation, it is determined that the legged robot's left straight punch cannot reach the opponent. This indicates that the swing trajectory of the fighting limb on the left side of the opponent does not intersect with the opponent's body shell. Therefore, the probability of hitting the opponent with a left straight punch is less than the probability threshold. In this embodiment, since the legged robot does not make any contact with the fighting opponent, it can be concluded that the probability of hitting the opponent is less than the probability threshold.

[0140] In the embodiments of this application, by determining the above three situations, the probability of the legged robot's attack action hitting can be accurately quantified, thereby deciding whether to perform orientation adjustment or displacement correction, so as to ensure that the final attack action has sufficient tactical effectiveness.

[0141] As an optional implementation, the fighting target includes a head, a torso, and multiple limbs connected to the torso. At least one memory is also encoded with additional instructions, which, when executed by at least one processor, cause the legged robot to: pre-calculate and evaluate the hit rate of multiple attack orientations, including: if the execution of an attack action can hit the head, torso, or at least one limb of the fighting target, determining that the probability of the attack action hitting the fighting target is greater than or equal to a probability threshold; if the execution of an attack action does not hit the head, torso, or multiple limbs of the fighting target, determining that the probability of the attack action hitting the fighting target is less than a probability threshold.

[0142] In this embodiment, if the attack action can hit the head, torso, or limbs of the fighting target, it indicates that the probability of the attack hitting the fighting target is greater than or equal to a probability threshold. If the attack action fails to hit the head, torso, or limbs of the fighting target, it can be determined that the probability of the attack hitting the fighting target is less than the probability threshold. This expands the hit area range of the legged robot when determining the hit probability of the attack action, extending the determination benchmark from the head and torso to include multiple limbs connected to the torso. This logic aims to more accurately assess the coverage of the overall effective hit area of ​​the fighting target by the attack action through more comprehensive geometrical overlap detection. The processor determines whether spatial contact can occur with the head, torso, or any limb of the fighting target by pre-calculating the swing trajectory of the attack action. Based on the detection results of the contact (first detection result and second detection result), the hit probability is divided into two intervals: "greater than or equal to the probability threshold" and "less than the probability threshold". The above-mentioned judgment logic is universal, applicable to both the first pre-calculation case based on the initial position and the second pre-calculation case based on the real-time dynamic position, thereby ensuring that the robot can make probabilistic judgments that conform to tactical logic in both the static prediction and dynamic tracking stages.

[0143] Optionally, during the determination process, the processor focuses on whether the attack action can effectively cover the effective hit area set of the fighting object. The effective hit area set may include the head, the object's torso, and multiple limbs connected to the object's torso. In the first pre-calculation case, pre-calculation is performed based on the initial position and the initial spatial position. If the first hit rate assessment result shows that the swing trajectory of the target attack limb or the end of the fighting limb executing the attack action by the legged robot spatially overlaps with any one of the fighting object's head, torso, or at least one limb located at the initial spatial position, the processor determines that the probability of the attack action hitting the fighting object is greater than or equal to a probability threshold. This indicates that, under the initial relative position (the relative situation between the initial position and the initial spatial position), the attack action has the possibility of hitting a valid part of the fighting object. In the second pre-calculation case, pre-calculation is performed based on the initial position and the fighting object's real-time current spatial position. If, during the adjustment of orientation or movement, the real-time pre-calculated swing trajectory spatially overlaps with the head, torso, or at least one limb at the current spatial position, it is similarly determined that the probability of the attack action hitting the fighting object is greater than or equal to a probability threshold.

[0144] Optionally, if the hit rate assessment shows that the attack action failed to cover the effective hit area of ​​the fighting target, it is determined to be a low-probability hit. In the first pre-calculation case, if the attack action fails to hit the fighting target's head, torso, or any limbs in the pre-calculation for the initial spatial position, i.e., the attack action deviates from the overall outline of the fighting target, then the probability of the attack action hitting the fighting target is determined to be less than the probability threshold. In the second pre-calculation case, if the attack trajectory has no spatial overlap with the fighting target's current head, torso, or any limbs in the dynamic real-time pre-calculation, the probability of the attack action hitting the fighting target is also determined to be less than the probability threshold. The above logic ensures that the invalidity of the attack can be identified in a timely manner when the attack action cannot reach any part of the fighting target's body. This helps the legged robot to actively adjust its strategy when a low-probability hit is determined, such as moving to a closer position, changing the attack angle, or stopping the current attack to avoid wasting energy and exposing defensive gaps.

[0145] In this embodiment, by expanding the judgment range to include the complete effective impact area, including the head, torso, and multiple limbs, the accuracy and robustness of the legged robot's hit probability determination are significantly improved. Compared to methods that only judge the torso and head, this approach more comprehensively reflects the hit situation in actual combat scenarios, especially providing finer-grained judgment criteria for changes in the impact area caused by the opponent's limb defense or movement. By being compatible with both the first pre-calculation (initial state) and the second pre-calculation (real-time state), reliable predictions can be provided not only before the attack is initiated but also real-time feedback corrections can be provided during execution. This effectively improves the legged robot's attack success rate, tactical adaptability, and operational efficiency in complex dynamic environments, while reducing the negative consequences of ineffective attacks.

[0146] For this embodiment Figure 4 Based on pre-calculation, it was determined that the swing trajectory of the legged robot's punch only made contact with the opponent's arm, without touching the opponent's core area. This indicates that the legged robot only touched the opponent's outer shell. In this embodiment, since the legged robot made contact with the opponent's arm, the probability of hitting the opponent is greater than or equal to a probability threshold.

[0147] As an optional implementation, the additional instructions, when executed by at least one processor, further enable the legged robot to: perform a hit rate assessment for multiple attack orientations through pre-calculation, including: if the minimum distance between the legged robot in its initial position and the fighting object is within a first distance range corresponding to the attack action to be performed, pre-calculate the process of performing the attack action under multiple attack orientations based on the initial position and the initial spatial position to obtain a hit rate assessment result, wherein the hit rate assessment result is used to represent the probability that the attack action hits the fighting object in the initial spatial position under multiple attack orientations; if the minimum distance between the legged robot in its initial position and the fighting object is within a second distance range corresponding to the attack action to be performed, perform the attack action in the initial position, wherein the minimum value of the second distance range is greater than the maximum value of the first distance range.

[0148] The aforementioned minimum distance can refer to the minimum distance among multiple points on the geometrically enclosing surface of the combatant, between the center of mass of the legged robot's torso.

[0149] Optionally, the aforementioned first distance range can be a pre-calculated close-range threshold interval based on the first combat command, used to determine whether high-precision hit pre-calculation is required. The upper limit (maximum value) of the aforementioned first distance range can be determined jointly by the minimum effective range of the attack action to be executed and the maximum extension capability of the attack end. The lower limit (minimum value) of the aforementioned first distance range can be zero or close to the contact distance. The aforementioned first distance range can be used to indicate that the legged robot and the combatant are in a critical hit zone, where the hit result of the attack action is highly dependent on the orientation and relative distance of the torso's forward face. For example, the first distance range of a straight punch can be 0.3 meters to 1.1 meters. Within the aforementioned first distance range, a 5° deviation in orientation can change the hit from striking the head to grazing the shoulder. The first distance range of a sweeping kick can be 0.5 meters to 1.2 meters. Whether the end trajectory can hit the leg or torso depends on the relative position of the combatant. If the distance between the initial spatial positions falls within the aforementioned first distance range, multi-directional or trajectory collision pre-calculation can be initiated to optimize the hit probability, thereby avoiding wasting computing resources at long distances and improving the response efficiency and energy consumption control of the legged robot.

[0150] Optionally, the aforementioned second distance range can be a long-distance threshold range preset according to the first combat command, where the attack action can be executed directly without pre-calculation. The lower limit (minimum value) of the aforementioned second distance range is greater than the maximum value of the first distance range. The upper limit (maximum value) of the aforementioned second distance range can be defined by the physical limits of the attack action, tactical rationality, and the size of the combat scenario (e.g., an octagon). The aforementioned second distance range can be used to indicate that the legged robot and the combatant are in an ultra-long-distance ineffective range, where even if the orientation or movement is adjusted, the end trajectory of the attack action cannot reach the geometric enclosure of the combatant. For example, if the attack action is a straight punch, the maximum effective radius of the straight punch is 1.2 meters, then the second distance range is more than 3 meters; if the attack action is a tornado kick, the maximum effective radius of the tornado kick is 1.8 meters, then the second distance range is more than 4 meters. That is to say, on the one hand, if the distance is within the second distance range, no pre-calculation process is required because the distance between the two is already far, the probability of hitting is 0, and pre-calculation is a waste of resources and meaningless. On the other hand, since the distance between the two is relatively far, it takes a long time for the legged robot to move and close the distance. In order to ensure timely response to combat commands and avoid users doubting the correctness of the legged robot's execution of combat commands, it can quickly respond to the first combat command and execute the corresponding attack action in place.

[0151] In summary, in this embodiment of the application, if the distance between the legged robot and the fighting object is within the aforementioned second distance range, no pre-calculation or movement is performed, and the attack action is directly triggered to maintain the integrity of the attack action rhythm, avoid delays in the user's fighting commands due to invalid pre-calculation, enhance the sense of ritual and visual continuity of the attack action, reduce processor load, and improve the overall real-time performance of the legged robot.

[0152] In this implementation, if the minimum distance between the legged robot in its initial position and the fighting target is within the first distance range corresponding to the attack action, the process of executing the attack action can be pre-calculated based on the initial spatial position and the initial position to obtain a hit rate assessment result. If the minimum distance between the legged robot in its initial position and the fighting target is within the second distance range, the attack action can be executed at the initial position. That is to say, after the legged robot receives the first fighting command, this embodiment can perform a hierarchical judgment on the spatial relationship between the current legged robot and the fighting target to decide whether to initiate high-precision hit pre-calculation or directly execute the attack action. The above process can achieve reasonable allocation of computing resources based on the distance range division, optimizing response efficiency and energy consumption while ensuring the effectiveness of the attack.

[0153] Optionally, in obtaining the minimum distance between the legged robot at its initial position and the fighting opponent, multiple distances in the world coordinate system can be obtained between the center of mass of the legged robot's torso and multiple points on the surface of the fighting opponent's geometric enclosure. The minimum value among these multiple distances can be determined as the minimum distance. For example, the minimum distance can be determined as the distance between the initial position coordinates and the position coordinates of the point closest to the initial position among the multiple points on the surface of the fighting opponent's geometric enclosure.

[0154] Optionally, based on the attack action defined by the first combat command, a first distance range and a second distance range corresponding to the attack action can be retrieved from a preset action parameter library. The first distance range is the valid judgment interval for the attack action. The upper limit of the first distance range can be determined by adding a safety tolerance to the maximum extension distance of the combat limb's extremity. This safety tolerance is a preset conservative allowance to address motion errors, visual positioning deviations, and the dynamic displacement of the combat target, ensuring that the attacking extremity can reliably hit the combat target even in its maximum extension state, preventing attack action failure due to minor deviations. For example, the safety tolerance can be preset to 5-10 centimeters. For instance, if the maximum extension distance of a straight punch is 1.2 meters and the safety tolerance is 30 centimeters, the first distance range can be 0-1.5 meters. The lower limit of the first distance range can be a close-range contact threshold, such as 0 meters. The second distance range can be a long-distance direct execution interval. The lower limit of the second distance range can be determined by adding a preset distance to the upper limit of the first distance range. The preset distance can be a limit that the user can tolerate for command response delay, or a limit that allows the legged robot to move according to the first combat command. The upper limit of the second distance range can be determined based on the physical limits of the action, tactical practicality, and the size of the combat scenario.

[0155] In this embodiment, the first and second distance ranges can be flexibly determined by the different attacks to be performed, ensuring clear boundaries for different limb lengths, joint ranges, and weapon configurations, rather than using a fixed distance range regardless of whether it's a kick or a punch.

[0156] Optionally, if it is determined that the minimum distance between the legged robot in the initial position and the fighting object is within the first distance range, the above probability pre-calculation process can be initiated, which will not be elaborated further here.

[0157] Optionally, if the minimum distance between the legged robot in its initial position and the fighting target is determined to be within the second distance range, the aforementioned probability pre-calculation process is not performed; instead, the attack action is directly triggered at the initial position. Whether the orientation of the torso's forward face needs to be adjusted before executing the attack action at the initial position depends on the specific instructions pre-set by the legged robot. For example, if the pre-set instruction for the minimum distance being within the second distance range is that the torso's forward face does not need to be adjusted, the attack action is executed directly at the initial position and orientation. This effectively reduces latency. Alternatively, if the pre-set instruction for the minimum distance being within the second distance range is that the torso's forward face needs to be adjusted, for example, adjusting the initial orientation to face the fighting target directly, this ensures the visual appeal of the fight. That is, even if the attack fails to hit the fighting target, the robot can still face the target, avoiding any misunderstanding of the legged robot's fighting process by not facing the target.

[0158] As an alternative implementation, at least one memory is also encoded with additional instructions that, when executed by at least one processor, cause the legged robot to: stop executing the current offensive action in response to receiving a combat action stop command sent by a remote control device, and switch to a defensive action, wherein the current offensive action is an attack action or a pure range attack action.

[0159] The combat action stop command refers to a control signal actively issued by the remote control device to suspend the currently executing offensive combat action and trigger a switch to defensive mode. This combat action stop command does not depend on whether the swing trajectory is completed or whether it hits the target, nor is it limited by the type of attack action (e.g., directional attack or pure area attack). Its core function is to provide users with emergency intervention capabilities to deal with scenarios such as sudden changes in battlefield situation, attack failure, or change of intent. This combat action stop command can be triggered by physical buttons, virtual buttons, voice commands, or motion-captured gestures, with the explicit semantic meaning of "immediately terminate the current attack behavior." The defensive action can be a passive response behavior performed by a legged robot to avoid or block the attack intent of a combatant, and can include non-aggressive limb movements such as blocking, crouching, jumping back, and posture deflection. Defensive actions also have spatial directionality.

[0160] Upon receiving the aforementioned command to stop the combat action, regardless of the current stage of the offensive action, the trajectory movement of the target's combat limbs is forcibly halted, and a preset defensive action is immediately initiated, such as adjusting the blocking posture, crouching to avoid an attack, jumping backward, or assuming a stable stance. The essence of this command is to establish a safety braking mechanism between the user's intent and the legged robot's behavior, preventing excessive attacks from wasting energy, causing postural imbalance, or exposing defensive openings. This ensures that the legged robot maintains tactical flexibility and survivability during combat, guaranteeing human-robot collaboration safety and controllable combat rhythm.

[0161] In this implementation, if a combat action stop command is received from the remote control device, the current offensive action can be stopped and the action can be switched to a defensive action. Figure 6 This is a schematic diagram of a fighting scene of a legged robot according to an embodiment of this application, such as... Figure 6 As shown, in this embodiment, during the execution of a combat mission, to ensure action safety, tactical flexibility, and real-time response capabilities in human-machine collaboration, if a combat action stop command is received, the legged robot terminates its current offensive action and switches to a preset defensive action (e.g., both hands are placed in a ready position, bent and positioned in front of the torso). This process does not depend on whether the swing trajectory is completed or whether it hits the target, nor does it distinguish between attack action types. The core logic is to achieve emergency braking and state reset between the user's intention and the legged robot's behavior, ensuring a rapid return to a safe and controllable defensive posture when the combat environment changes. Figure 7 This is a schematic diagram of a fighting scene of another legged robot according to an embodiment of this application, such as... Figure 7 As shown, if, during a defensive action, the first combat command is received instructing the legged robot to perform a front kick, the legged robot can lift its left leg and perform a front kick on the combat target.

[0162] Optionally, if the legged robot is executing an offensive fighting action triggered by a remote control device, the processor continuously loads and parses the instructions for the swing trajectory of the target fighting limb, driving the end effector of the corresponding fighting limb to move along a predetermined path. At this time, if the remote control device issues a stop command for the fighting action, the execution sequence of instructions for the current swing trajectory is immediately interrupted, and motion control signals are stopped from being sent to all joint drive units involved in the offensive action. Regardless of whether the attack action is in the initial arm raising, intermediate trajectory extension, or end effector contraction phase, all motion outputs are terminated, and the current joint position is locked to avoid inertial jitter. After the physical cessation of the attack action, a pre-stored defensive action is loaded. This program contains a set of standardized posture adjustment instructions, such as raising both arms forward to form a blocking posture, leaning the torso backward to lower the center of gravity, slightly squatting the legs to enhance stability, or performing a backward jump to create a safe distance. Each defensive action can be pre-designed based on the legged robot's body structure and kinematic constraints to ensure that posture switching can be completed in the shortest possible time without external interference and without causing new risks of imbalance.

[0163] As an optional implementation, at least one memory further encodes additional instructions, which, when executed by at least one processor, cause the legged robot to: , during the execution of a fighting action indicated by a current fighting command sent by a remote control device, respond to receiving a cancel fighting command sent by the remote control device, and within a time period after triggering the cancel fighting command, receive a next fighting command sent by the remote control device, stop executing the fighting action indicated by the current fighting command, and execute the fighting action indicated by the next fighting command; or, during the execution of a fighting action indicated by the current fighting command sent by the remote control device, respond to receiving a next fighting command sent by the remote control device, execute the fighting action indicated by the next fighting command after hitting a fighting target; or, during the execution of a fighting action indicated by the current fighting command sent by the remote control device, respond to receiving a next fighting command sent by the remote control device, execute the fighting action indicated by the next fighting command after completing the extended segment of the swing trajectory indicated by the current fighting command.

[0164] The aforementioned "cancel melee" command refers to a clear control signal actively issued by the remote control device to suspend the currently executing melee action and relinquish control for the next melee command. This "cancel melee" command is not the same as a melee action stop command; its semantics include not only terminating the current melee action but also implying the intention to "replace it with the melee action indicated by the next melee command." It serves as a pre-operation for tactical adjustments within a continuous combat sequence. This "cancel melee" command can be triggered via physical buttons, a virtual interface, or voice input. Its core function is to declare that the current melee action is no longer valid and that preparations should be made for the next response phase.

[0165] Optionally, the aforementioned duration range can refer to an acceptable time window set for receiving the next combat command after the combat command is canceled. This duration range can be measured in milliseconds, and its upper limit can be determined jointly by the legged robot's response latency, the inertial retention characteristics of its attack actions, and human-robot operating habits. This duration range aims to distinguish between "intention switching" and "misoperation," ensuring that unexpected actions are not mistakenly executed due to brief input jitter or delayed response. If no next combat command is received within this duration range, the canceled command will be invalidated by default, and the current combat action will continue.

[0166] Optionally, the next combat command can refer to a new combat command continuously issued by the remote control device to replace the current combat action. The priority of the next combat command is higher than the currently executing combat command. The swing trajectory can refer to the three-dimensional spatial motion path followed by the target combat limb from the initial posture to the completed posture, as defined by the combat command. The swing trajectory can be composed of a set of joint angles or end-effector pose nodes arranged in a time sequence, pre-modeled based on the kinematic model of the legged robot to ensure that the combat action conforms to physical feasibility and structural constraints. The unfolding segment can refer to the first half of the swing trajectory, from the initial posture to the end of the target combat limb reaching its maximum effective range or completing the main kinetic energy release phase. The unfolding segment can be a threatening, offensive, or defensive phase in an attack action, and the endpoint of the unfolding segment can correspond to the attack contact point or the peak velocity point of the trajectory. For example, in a "roundhouse kick" action, the unfolding segment is the stage from knee flexion to knee extension and the instep reaching maximum abduction displacement, while the convergence segment is the subsequent process of leg retraction.

[0167] In this implementation, if a cancel combat command is received during the execution of the combat action indicated by the current combat command triggered by the remote control device, and within the time range after the cancel command, the next combat command sent by the remote control device is received, the execution of the combat action indicated by the current combat command can be stopped, and the combat action indicated by the next combat command can be executed. If the next combat command is received from the remote control device, the combat action indicated by the next combat command can be executed after hitting the combat target using the combat action indicated by the current combat command. Alternatively, the combat action indicated by the next combat command can be executed after completing the unfolding segment of the swing trajectory indicated by the current combat command. That is to say, in this embodiment, in order to improve the smoothness of human-machine collaboration, the flexibility of tactical response, and the freedom of user operation during the execution of combat commands sent by the remote control device, three mutually exclusive but logically complete execution paths are constructed for the switching relationship between the "current combat command" and the "next combat command". The three paths correspond to three interaction modes: "rapid command replacement", "natural connection after hit" and "intelligent transition of action phase". They can ensure the complete expression of combat intentions and achieve high-precision transmission of user tactical intentions and accurate response of robot behavior without destroying the physical continuity of actions.

[0168] Optionally, during the execution of the combat action indicated by the current combat command, if the remote control device triggers a cancel combat command and receives the next combat command within the set time range, it will immediately terminate the trajectory execution of the current combat action and directly load and start the motion control sequence corresponding to the next combat command. For example, if the user finds that the target has dodged when executing a "right straight punch", they can immediately press the cancel button and then trigger the "left hook" command. After confirming that the command sequence meets the time constraint, the straight punch action will be stopped, and the hook punch trajectory will be executed immediately. This mechanism allows users to perform "cancel and reselect" combo operations in high-speed combat, greatly improving operational efficiency and tactical diversity, avoiding missed attack opportunities due to having to wait for the action to end, and significantly enhancing the real-time response capability of combat and the realism of human-computer interaction.

[0169] Optionally, during the execution of the combat action indicated by the current combat command, if the remote control device sends the next combat command directly without triggering a cancel combat command, the current action will not be interrupted immediately. Instead, the current trajectory will be terminated and the next combat command will be initiated only after the target limb of the current combat action successfully hits the target combat object. The hit determination is based on the contact signal fed back by the visual perception module, the impact peak detected by the torque sensor, or the collision detection result between the preset trajectory and the combat object. For example, if the user issues the "upward kick" command before the "sweeping leg" is completed, the upward kick action will only naturally transition after the sweeping leg hits the opponent's leg or torso. This mode ensures that each combat action can complete its core attack intention, preventing attacks from becoming ineffective or wasting energy due to interruption, while maintaining the continuity of the attack rhythm, giving the combo process realistic physical feedback and tactical rhythm, and enhancing the immersion and rationality of the combat experience.

[0170] Optionally, during the execution of the combat action indicated by the current combat command, if the remote control device triggers the next combat command without sending a cancellation command, it can also be selected to immediately switch to the movement sequence corresponding to the next combat command after the completion of the unfolding segment of the current swing trajectory. The unfolding segment refers to the first half of the movement range in the swing trajectory from the starting posture to the end of the target combat limb reaching its maximum effective range or releasing the main kinetic energy. Its endpoint is determined by the peak kinetic energy of the trajectory node, the rate of change of the end velocity, or a preset marker point. For example, when executing a "roundhouse kick," once the leg completes the unfolding segment from flexion to extension, it is determined that the core attack energy has been released. At this time, even if the leg retraction action has not yet been completed, the "downward elbow strike" command can be initiated in advance. This mechanism shortens the command waiting delay without sacrificing the effectiveness of the attack, making the tactical combination more compact and the rhythm faster. It is especially suitable for high-frequency, multi-dimensional combination attack scenarios, achieving efficient connection of "new move started before the action is finished," and improving tactical adaptability and operational smoothness under high combat intensity.

[0171] In this embodiment, the three switching mechanisms described above together constitute a multi-layered, multi-scenario combat command response system, serving three typical combat requirements: "rapid change of intent," "attack must be completed," and "flexible acceleration of pace." Through semantic recognition and state awareness, the system intelligently matches the user's operational intent with the robot's execution capabilities, ensuring that combat behavior maintains both physical realism and attack effectiveness while possessing a high degree of human-machine collaborative flexibility.

[0172] As an alternative implementation, at least one memory is also encoded with additional instructions, which, when executed by at least one processor, cause the legged robot to: in response to receiving a set of fighting instructions sent by a remote control device, parse multiple fighting instructions to be executed from the set of fighting instructions; and sequentially execute the fighting actions indicated by the fighting instructions in the direction of the initial spatial position according to the chronological order in which the multiple fighting instructions were received.

[0173] The combat instruction set refers to a structured control data packet sent at once by a remote control device, containing multiple independent combat instructions. Essentially, it's a sequence of combat actions encapsulated according to a preset protocol and with a clear temporal relationship. This combat instruction set can also be called an instruction package, consisting of multiple complete combat instructions. Each instruction independently defines the action type, the method of selecting the combat target, the scope of effect, and execution parameters, such as "right straight punch—left sweep kick—uppercutive punch." The combat instruction set does not rely on the immediate response of individual instructions but is received and parsed in batches by the legged robot as a whole. Its design aims to support users in pre-setting compound tactical actions in a single operation, enabling the efficient delivery of combos, combination techniques, or predictive attack sequences. This combat instruction set can be selected in batches by the user through the remote control interface, invoked through preset skill templates, or generated after recording by motion capture equipment.

[0174] Optionally, the aforementioned time sequence can refer to the linear arrangement of multiple combat commands in the combat command set, formed by the time order (sequential order) of their transmission from the remote control device to the communication module of the legged robot. This time sequence can be precisely recorded and sorted based on the timestamps of command reception, ensuring that each subsequent command is generated after the preceding command has been confirmed as received, and is not disrupted by network latency, command length differences, or transmission priorities. The determination of the time sequence does not depend on the logical correlation of the command content, but is solely based on the reception sequence. For example, if the "straight punch" command is received first, followed by the "sweeping kick" command, then regardless of whether the two are tactically reasonable, they will be executed in the "straight punch—sweeping kick" order. This mechanism ensures the original integrity of the user's operational intent, avoids autonomous rearrangement or merging of combat commands, and ensures that the execution result of each tactical combination is consistent with the user's preset.

[0175] In this implementation, if a set of combat instructions is received from a remote control device, multiple combat instructions to be executed can be parsed from the set. The combat actions indicated by these instructions can be executed sequentially according to the time order in which they are received. In other words, in this embodiment, when the legged robot executes complex combat tactics, to support the user in issuing multi-step attack sequences with a single operation, it receives the set of combat instructions sent by the remote control device through a communication module and completes instruction parsing and sequential execution according to a structured format. This process does not rely on the user sending instructions one by one; instead, it receives, parses, and executes the instructions in a complete instruction package according to a strict time sequence, achieving a leap from "single instruction response" to "multi-action orchestration," providing a reliable execution foundation for the implementation of advanced combat strategies.

[0176] Optionally, if the legged robot receives a combat instruction set from a remote control device via its communication module, it initiates instruction parsing to perform protocol decoding and structure verification on the combat instruction set. The combat instruction set is a structured data unit encapsulating multiple independent combat instructions. Each combat instruction includes information such as action type, selection method of the combat target, range of action, and execution parameters. Based on a pre-set communication protocol, each combat instruction is identified and separated, converted into an internally executable control instruction object, and each control instruction object is bound to the original receiving sequence number and timestamp in the instruction set to ensure the integrity and traceability of the instruction content.

[0177] Optionally, after instruction parsing is complete, an execution queue can be established based on the order in which each combat instruction is received in the combat instruction set, i.e., the chronological order. This chronological order is determined by the internally recorded instruction reception timestamps. Following the order of each combat instruction in the execution queue, the corresponding combat action indicated by that instruction is initiated one by one. Once the previous combat action is completed or aborted, the action defined by the next instruction is automatically loaded and executed. For example, if the combat instruction set contains three instructions: "right straight punch—left sweep kick—uppercut," they will be executed in the above order. Even if the combatant has moved while executing the sweep kick, the uppercut will continue to be executed without intention re-judgment or order reordering. This mechanism ensures that the user's preset tactical arrangement is completely restored, achieving "how the user wants to fight, the robot executes," avoiding tactical distortion caused by unauthorized intervention or optimization.

[0178] As an alternative implementation, the additional instructions, when executed by at least one processor, further enable the legged robot to: in the process of executing a fighting action indicated by the current fighting instruction in the fighting instruction set, in response to hitting a fighting target, execute, in chronological order, the fighting action indicated by the next fighting instruction of the current fighting instruction in the direction of the initial spatial position; or, in the process of executing a fighting action indicated by the current fighting instruction in the fighting instruction set, in response to completing the unfolding segment of the swing trajectory indicated by the current fighting instruction, execute, in chronological order, the fighting action indicated by the next fighting instruction of the current fighting instruction in the direction of the initial spatial position.

[0179] In this implementation, during the execution of the combat action indicated by the current combat instruction in the combat instruction set, if the aforementioned combat action hits the combat target, the next combat action indicated by the next combat instruction can be executed in chronological order. Alternatively, after completing the expansion phase of the combat instruction indicated by the current combat instruction, the next combat action can be executed in chronological order. In other words, in this embodiment, during the execution of the combat instruction set sent by the remote control device, to achieve natural connection of tactical combos and efficient transmission of attack effectiveness, the legged robot automatically initiates the execution process of the next combat instruction based on two selectable triggering conditions during the execution phase of the current combat instruction, such as hitting the combat target or completing the expansion phase of the swing trajectory. This mechanism does not rely on user-initiated triggering and does not interrupt the physical integrity of the current combat action. Instead, through the legged robot's state perception and end-effector trajectory analysis, it intelligently advances the instruction sequence when specific tactical nodes are met, achieving a smooth combat rhythm of "attack as connection, hit as continuation," significantly enhancing the practicality of combat strategies and the immersive experience of operation.

[0180] Optionally, during the execution of the combat action indicated by the current combat instruction in the combat instruction set, if the visual perception module or torque sensor determines that the end of the target limb has made effective contact with the target opponent, i.e., a hit is determined, the current action's convergence phase is immediately terminated. Instead of waiting for a complete end, the robot directly loads and executes the combat action corresponding to the next combat instruction according to the time sequence of the combat instruction set. For example, when executing a "sweeping leg" action, if its foot detects physical contact with the opponent's leg or torso, the attack is deemed effective, and the robot immediately jumps to the preset next instruction, "uppercut," without requiring secondary user input. This mechanism uses the "effectiveness" of the attack as the trigger for instruction progression, directly binding combo behavior to actual combat results. This avoids rhythmic breaks or energy waste caused by "missed attack transitions," significantly improving the tactical efficiency and realism of the attack sequence, making the legged robot's combat behavior closer to the coherent logic of "hit-based" combat in real combat.

[0181] Optionally, during the execution of the combat action indicated by the current combat command in the combat command set, if no hit event is detected, but the completion of the current swing trajectory's unfolding phase has been determined, the execution of the next combat command can also be initiated. For example, when executing a "spinning kick," once the leg completes the acceleration phase from flexion to full extension, it is determined that the core attack energy has been released. Even if the target is not hit, it is still considered that the action has fulfilled its tactical intent, and the execution switches to the next command, "downward elbow strike." This mechanism breaks through the rigid constraint of "must hit to connect," allowing tactical advancement based on the completion of the action while ensuring that the attack action is not interrupted prematurely. This avoids command stagnation and combat blockage caused by the opponent's evasion, enhances adaptability and continuity in dynamic combat, and makes combo actions have a wider range of applicable scenarios and higher fault tolerance.

[0182] In this embodiment, the two triggering mechanisms described above together constitute an intelligent combo advancement system oriented towards practical combat needs. They are based on "attack result orientation" and "action efficiency orientation," respectively, forming complementary execution logics. The first mechanism ensures that each step is built on effective strikes, enhancing the pressure and realism of the attack. The second mechanism maintains the combat rhythm even when there are no hits, avoiding tactical gaps due to environmental uncertainties. Both are executed sequentially based on the combat command set, without introducing autonomous decision-making, ensuring the complete reproduction of the user-preset tactical sequence. This process elevates the legged robot from "passively responding to commands" to "actively sensing the state and intelligently advancing tactics." Without sacrificing user control, it achieves automated connection and rhythm optimization of combat actions, greatly improving the smoothness of multi-command combinations, the coherence of combat performance, and the immersive experience of human-machine collaboration. This constructs a practically valuable intelligent execution paradigm for advanced combat robots.

[0183] As an optional implementation, the combat limb endpiece includes at least one of the following: in response to the legged robot's limb endpiece holding a combat device, the combat limb endpiece is the attack end of the combat device, wherein the combat device includes a first type of combat device and a second type of combat device, the attack end of the first type of combat device is the attack endpiece of the first type of combat device, and the attack endpiece of the second type of combat device is the second type of combat device itself; in response to the limb endpiece not holding a combat device, the combat limb endpiece is a limb endpiece.

[0184] In this implementation, if the endcaps of the legged robot are equipped with a fighting weapon, the attacking end of the fighting weapon can be identified as the endcap of the fighting limb. Conversely, if the endcaps are not equipped with a fighting weapon, the endcaps can be identified as the endcap of the fighting limb.

[0185] Combat implements can refer to external physical devices held by the end of a legged robot's limbs to enhance or extend its attack capabilities. Their function is to expand the attack range, increase the intensity of the strike, or change the attack pattern, achieving an attack effect that is not achieved by the human limb itself. Combat implements can be rigid structures or deformable devices, and can include, but are not limited to, hammers, clubs, blades, spears, and gauntlets. Essentially, they are the force transmission medium or contact medium when the robot performs combat actions.

[0186] Combat equipment can be categorized into Type I and Type II combat equipment. Type I combat equipment refers to equipment whose attack point is clearly distinct from its main body structure, possessing an independent attack surface or concentrated point of impact; its attack effectiveness depends on physical contact at specific points. The attack point of Type I combat equipment is the end structure inherent to the equipment itself, specifically designed for launching an attack; it can be sharp, hard, or a region of concentrated force. Examples include the head of a hammer, the blade of a knife, and the tip of a spear. The design principle of Type I combat equipment is to achieve penetration, striking, or cutting effects through localized high-stress points; its attack effectiveness highly depends on the spatial positioning and trajectory accuracy of the attack point. Type II combat equipment refers to equipment whose entire structure possesses attack capabilities, without a clear distinction between attack and non-attack points; its attack effectiveness is achieved through all or most of the surface of the equipment body, with the attack point being the equipment itself. The second type of combat weapon mentioned above can be a block, rod, or ring-shaped structure without sharp points or concentrated force, such as an iron rod, metal arm armor, a circular hammer, or a long whip. The attack method depends on the overall mass, swing inertia, or coverage area, rather than a single point of contact. For example, a metal rod of uniform length can produce an effective strike no matter where it hits the combatant; therefore, the attacking end of the aforementioned metal rod is the entire body of the metal rod.

[0187] Optionally, the aforementioned attack endpoint can refer to the physical end point or minimum effective area on a Type I combat implement specifically designed for launching an attack and in direct contact with the combat target. It can be the coordinate position with offensive effectiveness within the swing trajectory of the combat implement. The attack endpoint is a reference point used to calculate the trajectory, determine hit rate, and assess the contact state during combat maneuvers. For example, in a hammer, the attack endpoint is the striking surface at the very tip of the hammerhead; in a spear, the attack endpoint is the sharpest point of the spearhead. The aforementioned attack endpoint can also refer to the physical area on a combat implement used to interact with the combat target and produce an attack effect. It can be a reference position used to determine the effectiveness of an action and whether it has hit the target when executing combat commands. The definition of the attack endpoint varies depending on the type of combat implement: in Type I combat implements, the attack endpoint is the attack end point; in Type II combat implements, the attack endpoint is the entirety of the Type II combat implement.

[0188] As an optional implementation, the types of combat commands include at least one of the following: command types triggered by physical buttons on a remote control device, command types triggered by virtual buttons on a remote control device, voice commands, and motion capture commands triggered by motion capture devices.

[0189] In this implementation, the types of combat commands can include command types triggered by physical buttons on the remote control device, command types triggered by virtual buttons on the remote control device, voice types, or motion capture types triggered by motion capture devices.

[0190] Among these, commands triggered by physical buttons refer to combat control commands triggered by physical hardware components such as buttons, switches, or levers on the remote control device, which have a clear mechanical structure and tactile feedback. These commands send preset combat action signals to the legged robot by the user directly pressing, moving, or switching physical components on the remote control device. Physical buttons can correspond to single actions, such as a "right straight punch button" or a "sweeping kick button," or they can be combination buttons, such as a long press of "attack + direction" to trigger a specific trajectory. These commands triggered by physical buttons do not rely on screen displays or external perception, making them suitable for high-speed combat, scenarios with limited visibility, or scenarios requiring blind operation.

[0191] Optionally, the command type triggered by the aforementioned virtual buttons can refer to combat control commands triggered by interactive controls without physical structure, presented in a graphical interface on the remote control device's display screen. This command type can activate the corresponding combat action transmission protocol by the user touching icons, buttons, or areas on the screen, responding to the collaborative work of the touch sensor and graphics rendering. Virtual buttons can dynamically change their layout, shape, color, or labels, supporting multi-level menus, skill tree switching, or skinned interfaces; for example, displaying clickable icons such as "Combo 1" or "Area Sweep" on the screen. The command type triggered by the aforementioned virtual buttons is suitable for scenarios requiring a rich action library, customizable interface, or tactical preview. Its advantages lie in high space utilization and strong scalability, allowing users to add new commands through software updates.

[0192] Optionally, the aforementioned voice type can refer to a control method where the user issues natural language commands to the remote control device via a voice input device, which are then parsed into structured combat commands by a voice recognition module. This voice type can use human language as a medium, such as "right straight punch," "left hook," or "launch an area attack," and converts the voice content into combat actions that the robot can execute through voiceprint recognition, semantic understanding, and command mapping algorithms. This voice type does not rely on physical contact or visual interaction and is suitable for scenarios requiring both hands, remote control, or covert operation. Its core value lies in lowering the operational threshold and improving the naturalness and intuitiveness of command input. The aforementioned motion capture type can refer to a control method where external motion capture devices collect user limb movement data, which is then converted into robot combat commands through spatial coordinate mapping and motion recognition algorithms. This motion capture type can rely on wearable sensors, optical capture, or inertial measurement units to record the user's movement trajectory in real space, such as arm swings, leg raises, and turns. The captured movement trajectory is then matched with a preset combat action template to generate corresponding combat actions to be executed by the legged robot. For example, when a user makes a "right straight punch" motion, the motion capture device captures the relative motion vectors of the shoulder and elbow, identifies them, and triggers the robot to execute the same type of motion. This motion capture method does not rely on buttons or voice, but uses the user's own movements as the control source, achieving high-fidelity reproduction of the movements. It is suitable for pre-calculation training, immersive combat, or professional-grade control scenarios.

[0193] Figure 8 This is a schematic diagram of another legged robot according to an embodiment of this application, such as... Figure 8 As shown, the legged robot 80 includes a torso 81; multiple target fighting limbs 82 and corresponding multiple fighting limb ends 83 respectively connected to the torso; at least one visual perception module 84 configured to acquire visual information of the surrounding environment of the legged robot, the visual information of the surrounding environment including at least one fighting object; at least one communication module 85 configured to receive movement commands or fighting commands sent by a remote control device; at least one processor 86; and at least one memory 87, the at least one memory encoding instructions, which, when executed by the at least one processor, cause the legged robot to: step S802, move in the target direction indicated by the received movement command sent by the remote control device; step S804, execute the pure range attack action indicated by the second fighting command, which includes target pure range attack action information, in response to the received fighting command sent by the remote control device.

[0194] Regarding the aforementioned hardware, and Figure 1 The functions are similar, so they will not be described in detail here.

[0195] In this embodiment, at least one memory 87 is encoded with instructions that, when executed by at least one processor, cause the legged robot to perform step S802: in response to a movement instruction sent by a received remote control device, move in the target direction indicated by the movement instruction.

[0196] Regarding the execution process of the aforementioned movement commands, and Figure 1 The steps in step S102 are the same as those in the previous section, and will not be described in detail here.

[0197] In this embodiment of the application, at least one memory 87 is encoded with instructions that, when executed by at least one processor, cause the legged robot to perform step S804: responding to a received fighting instruction sent by a remote control device as a second fighting instruction, the second fighting instruction including target pure range attack action information, and performing the pure range attack action indicated by the second fighting instruction.

[0198] The aforementioned second combat command can be a control command sent by a remote control device to trigger the legged robot to perform omnidirectional combat behavior. Its characteristic is that it does not rely on the relative orientation or spatial alignment between the legged robot and the combatant, but is triggered by determining whether the combatant is within the action range of a pure area-of-effect attack. The command content of the aforementioned second combat command can include an action type identifier and an action range parameter. The action type identifier can include sweeping kicks, 360° rotating kicks, torso rotation swings, omnidirectional ground-shaking impacts, etc. The swing trajectory of the aforementioned pure area-of-effect attack can be circular, fan-shaped, or omnidirectionally diffused. The effective action area is defined by the length of the target combat limb of the legged robot and its radius of motion. The execution of the second combat command is not based on "hitting a specific target," but rather on "whether the combatant falls within the corresponding action range."

[0199] The command content of the aforementioned second combat command may include information about the target's pure area-of-effect attack actions. This information may include a type identifier for the pure area-of-effect attack action.

[0200] The aforementioned pure range attack actions refer to the all-area combat behavior of a legged robot that can be effective without relying on the orientation of its forward face or the relative position of the combatant. The attack range of these pure range attack actions covers a large spatial area centered on the legged robot, and the trajectory of these pure range attack actions does not have a clear directional preference or locking attribute. In short, compared to attack actions other than pure range attacks, pure range attack actions do not consider orientation adjustments; that is, the effectiveness of the execution of these pure range attack actions does not depend on the relative orientation relationship between the legged robot and the combatant, but rather on the combatant being within the physical space covered by the trajectory of the pure range attack action for a valid hit. Pure range attack actions can include, but are not limited to: sweeping kicks, 360° rotating kicks, omnidirectional swings, and torso rotation strikes. The essential characteristic of these pure range attack actions is that the action domain is a non-directional geometric range, and "alignment" is not a prerequisite for execution. These pure range attack actions can also be called non-directional attack actions, all-area coverage attacks, or all-area attack actions.

[0201] In this embodiment, if the legged robot receives a second combat command from the remote control device, it will initiate a response process that focuses on spatial coverage and requires no posture calibration—a pure range attack. This response process can include three stages: acquiring the legged robot's own spatial reference, identifying the spatial position of the combat target, and directly executing a preset pure range attack action. Using a world coordinate system as a unified reference framework, the processor, in collaboration with the visual perception module and motion control module, enables rapid deployment and effectiveness verification of the attack range without relying on adjustments to the forward orientation of the torso, thereby improving attack efficiency and response speed in complex, dynamic combat environments.

[0202] Optionally, in response to the received fighting command sent by the remote control device as a second fighting command, the initial position of the legged robot is obtained, and the initial spatial position is obtained from the visual information; the pure range attack action indicated by the second fighting command is executed, wherein obtaining the initial spatial position from the visual information is only one method for obtaining the position of the fighting object in the environment of the fighting scene, and no specific limitation is made here.

[0203] Optionally, during the stage of obtaining the legged robot's own spatial reference, the processor can read the initial position of its own torso from the inertial measurement unit and joint encoder of the legged robot at the moment it receives the second fighting command. For example, the spatial coordinates of the center of mass of the legged robot's torso in the world coordinate system can be determined as the initial position.

[0204] Optionally, during the stage of identifying the spatial position of the fighting object, the processor can invoke the visual perception module to collect visual information of the current environment, and locate the initial spatial position of at least one fighting object in the world coordinate system through target recognition algorithm and three-dimensional pose estimation algorithm. For example, the initial spatial position can be determined by the spatial coordinates of the center of mass of the fighting object's torso at the moment the second fighting command is triggered.

[0205] Optionally, during the stage of directly executing the preset pure range attack action, the processor can directly drive the legged robot to execute the corresponding pure range attack action based on the action type and range parameters carried by the second combat command. For example, if the second combat command is a sweeping leg kick, the processor can generate a circular or fan-shaped attack area centered on the initial position and within a set radius, based on the leg length, rotation radius, and swing trajectory model of the legged robot, and detect whether the initial spatial position is within the attack area. If the combat target is within the attack area, the leg swing action is initiated, causing the foot of the legged robot to complete the sweeping attack along the preset trajectory without any posture correction; if the combat target is outside the attack area, the range attack action is still executed, but the attack is ineffective, and no movement or redirection is performed.

[0206] In this embodiment, the essential difference in execution flow between the first and second combat commands stems from the different physical characteristics and spatial dependencies of the combat actions they indicate. The first combat command indicates attack actions other than pure range attacks, with a localized, directional, and trajectory-limited attack range, such as a straight punch or hook. Therefore, it is necessary to obtain the initial orientation, calculate the first target orientation, and adjust the orientation to ensure that the swing trajectory of the combat limb's end-effector covers the combat target. The second combat command, on the other hand, indicates pure range attacks, with a global, direction-unrestricted geometric area, such as a sweeping kick or a 360° spinning kick. The attack effectiveness is determined by spatial coverage and does not rely on "alignment." Therefore, it does not require obtaining the initial orientation or calculating and adjusting to the first target orientation. Execution can be triggered by determining whether the combat target falls within the attack range (attack area) corresponding to the pure range attack. These differences reflect the design principles of the two commands: the first combat command aims for precise hits, while the second combat command aims for area suppression.

[0207] In this embodiment, the method directly determines whether the target falls within the action domain of a pure range attack action based on the initial position and initial spatial position under the second combat command, and directly executes a direction-independent global attack action. This avoids the necessary forward orientation adjustment and trajectory pre-calculation process required for pure range attack actions, significantly reducing the response latency and computational overhead of the legged robot. This mechanism effectively matches the essential characteristics of pure range attack actions such as sweeping kicks and 360° rotating kicks, achieving attacks through spatial coverage rather than orientation alignment. This allows the legged robot to efficiently launch attacks even in environments with multiple targets, close range, or dynamic interference, improving the suddenness, coverage, and robustness of the attack. Simultaneously, it reduces the "missed punches" phenomenon caused by orientation adjustment failures, effectively enhancing the tactical practicality and stable performance of the legged robot in actual combat.

[0208] The legged robot described above in the embodiments of this application will be explained in further detail below.

[0209] As an alternative implementation, at least one memory is further encoded with additional instructions, which, when executed by at least one processor, cause the legged robot to: respond to a received combat instruction sent by a remote control device as a second combat instruction, and execute a pure range attack action indicated by the second combat instruction, including: responding to a received combat instruction sent by a remote control device as a second combat instruction, and after obtaining the initial position of the torso in the world coordinate system and the initial spatial position of the combat object in the world coordinate system, executing the pure range attack action indicated by the second combat instruction.

[0210] In this embodiment, when a combat command (the second combat command) is received from the remote control device and a pure area-of-effect attack is executed, if the second combat command is received, the initial position of the torso in the world coordinate system and the initial spatial position of the combat target in the world coordinate system can be obtained. After obtaining the aforementioned initial position and initial spatial position, the pure area-of-effect attack can be executed.

[0211] Optionally, during the process of obtaining the initial position of the torso, the processor can determine the coordinates of the geometric center or center of mass of the legged robot's torso in the world coordinate system using the inertial measurement unit and kinematic model inside the legged robot, combined with the absolute positioning information provided by the visual perception module. These coordinates can then be used as the initial position of the torso.

[0212] Optionally, during the process of acquiring the initial spatial position of the fighting object, the processor can utilize visual information captured by the visual perception module to determine the target object of the current interaction from at least one fighting object using a target recognition algorithm. The processor can calculate the spatial extent of the geometric bounding volume of the aforementioned fighting object in the world coordinate system and extract the coordinates of each point on the surface of the aforementioned geometric bounding volume, thereby determining the initial spatial position of the fighting object.

[0213] Optionally, after the processor successfully obtains the initial position of the torso and the initial spatial position of the combat target, there is no need to make fine adjustments to the torso orientation based on the relative angle between the two, nor is there a need to perform complex pre-calculation of the hit probability. The processor can directly send control commands to the drive module of the legged robot, driving the target attack limb (e.g., leg or arm) to move along a fixed trajectory preset by the second combat command, executing a pure range attack action.

[0214] In this embodiment, the method described above significantly reduces the computational load and control complexity of legged robots when performing pure range attack actions by simplifying the execution logic of pure range attack actions. Compared to targeted attacks that require high-precision orientation adjustment and real-time hit assessment, pure range attack actions rely on predefined motion trajectories and spatial coverage areas. As long as the combat target is within the effective attack radius, a strike can be inflicted, without the need for a complex real-time decision-making process.

[0215] As an optional implementation, at least one memory further encodes additional instructions, which, when executed by at least one processor, cause the legged robot to: execute a pure range attack action indicated by a received combat instruction (a second combat instruction) sent by a remote control device, including: if the minimum distance between the torso at the initial position in the world coordinate system and the combat target is within a first attack range indicated by the second combat instruction, execute the pure range attack action at the initial position; if the minimum distance between the torso at the initial position and the combat target is within a second attack range indicated by the second combat instruction, move from the initial position to a second target position, and execute the pure range attack action at the second target position, wherein, when the torso is at the second target position, the combat target is within the first attack range, and the minimum value of the second attack range is greater than the maximum value of the first attack range; if the minimum distance between the torso at the initial position and the combat target is within a third attack range indicated by the second combat instruction, execute the pure range attack action at the initial position, wherein the minimum value of the third attack range is greater than the maximum value of the second attack range.

[0216] The first attack range refers to a pre-defined range of action distances that allows the torso to launch an attack directly from its initial position without displacement when performing a pure range attack. The upper limit of this first attack range can be determined by the maximum effective radius of the target limb (e.g., leg or arm) during the attack. The lower limit of this first attack range can be defined by the minimum safe distance required for the initial action of the attack to contact the target. For example, when the attack is a sweeping leg kick with an effective attack radius of 1.2 meters at the end of the leg, the first attack range can be defined as 0.3 to 1.5 meters. Here, 0.3 meters is the minimum effective distance to avoid contact with the target before the attack is fully deployed, and 1.5 meters is the maximum effective boundary to ensure the swing trajectory completely covers the torso or limbs of the target. Within this first attack range, the legged robot can ensure effective intersection between its swing trajectory and the target without adjusting its position, achieving an "instant hit" attack effect. The aforementioned first attack range setting aims to maximize attack efficiency and response speed, reduce unnecessary movement delay, and is suitable for combat scenarios where the fighting opponent is within the effective range of the target's fighting limbs and the space allows for stable execution of actions.

[0217] Optionally, the aforementioned second attack range can refer to the intermediate distance range between the legged robot and the fighting target after receiving a pure range attack command, where the distance exceeds the first attack range but is still within the compensable movement range. The lower limit of the aforementioned second attack range can be the maximum value of the first attack range, and the upper limit of the aforementioned second attack range can be the maximum movable compensation boundary set based on the legged robot's mobility, attack delay tolerance, and the dynamic characteristics of the fighting target. For example, when the upper limit of the first attack range is 1.5 meters, and the fighting target is 1.6 meters to 2.2 meters away from the legged robot, this distance is determined to be within the second attack range. In this case, instead of directly executing an attack, a short displacement path is planned from the initial position to the second target position, so that the robot's initial spatial position falls within the first attack range after moving to the second target position. In the embodiments of this application, by defining the aforementioned second attack range, the aim is to make attacks that were originally "barely reachable" feasible again by fine-tuning the position, avoiding attack failures caused by being "too close to hit" or "too far to be effective." The design of the second attack range reflects the intention to proactively optimize attack effectiveness. The core is to "exchange the minimum displacement for the maximum hit guarantee", which enhances the adaptability and accuracy of the legged robot under edge distance conditions while maintaining the attack rhythm.

[0218] Optionally, the third attack range can refer to the long-distance interval determined as "uncompensable distance" after the distance between the fighting target and the legged robot exceeds the upper limit of the second attack range. The lower limit of the aforementioned third attack range can be the maximum value of the second attack range. The upper limit of the aforementioned third attack range is theoretically not strictly limited, but in practical applications it can be constrained by factors such as the timeliness of the attack action, the movement speed of the fighting target, and the response delay of the legged robot. For example, if the distance between the legged robot and the fighting target exceeds 2.5 meters, it is determined that the distance is within the third attack range. Within this range, if the legged robot attempts to move into the effective attack range of the fighting target, the attack cannot be guaranteed to hit due to the excessive movement time (e.g., exceeding 0.8 seconds), as the fighting target may have already shifted, turned, or deviated from the swing trajectory during this period. Therefore, abandoning the displacement strategy and directly triggering a pure range attack action at the initial position, so that the swing trajectory is released in a wide-area coverage, although it cannot guarantee a precise hit, it can create a deterrent or area suppression effect. The setting of the third attack range in this application embodiment is not for the purpose of hitting the target, but to maintain the continuity of the attack behavior and the tactical presence, and to prevent the user's perception from being interrupted due to "no attack". Its function is to balance the expression of the attack intention and the rational use of resources, maintain the integrity of the combat rhythm in long-range confrontation, and avoid the sense of frustration caused by no response.

[0219] Optionally, the aforementioned second target position can be the execution position that the legged robot actively plans and moves to when it receives the second combat command and is within the second attack range, in order to ensure the effective hit of the pure range attack action. This execution position is not arbitrarily selected, but rather a feasible target coordinate calculated from the initial position using the shortest path, based on the initial spatial position and the effective radius of the attack action, provided that the distance is within the first attack range. The core function of this second target position is to use minute displacement compensation to bring the combat target, which was originally outside the direct attack range, back into the swing trajectory coverage area, achieving precise control of "ineffective if stationary, hit if slightly moved," avoiding attack failure due to blind movement or delayed execution.

[0220] In this implementation, if the minimum distance between the torso at its initial position in the world coordinate system and the combat target is within the first attack range, a pure range attack action can be performed at the initial position. If the minimum distance between the torso at its initial position in the world coordinate system and the combat target is within the second attack range, it can move from the initial position to the second target position and perform a pure range attack action at the second target position. If the distance is within the third attack range, a pure range attack action can be performed directly at the initial position. That is, in this embodiment, when the legged robot executes the pure range attack action indicated by the second combat command, it dynamically divides and responds to three mutually exclusive distance intervals based on the minimum distance: the first attack range, the second attack range, and the third attack range. The above mechanism is not simply a judgment of whether it can hit, but rather a three-level response strategy constructed based on the physical characteristics of the attack action, the dynamics of the target, and the consistency of user experience. It realizes a layered execution logic of "precise hit - controllable compensation - tactical deterrence," ensuring that the pure range attack has reasonable, stable, and expected response behavior under different spatial relationships.

[0221] Optionally, if the minimum distance between the torso at the initial position in the world coordinate system and the combat target is within the first attack range corresponding to the second combat command, the combat target is determined to be within the effective radius of the pure range attack action, and the swing trajectory can be ensured to completely cover the combat target's body without displacement. The attack action is initiated directly from the initial position without any position adjustment. For example, when the attack action is a sweeping leg kick with an effective radius of 1.2 meters, and the combat target's torso is within 1.5 meters, this distance is determined to be within the first attack range, and the complete sweeping leg trajectory is immediately executed, causing the target's limbs to sweep the combat target along a preset path. This response mode has the highest efficiency and the lowest latency, and is suitable for typical close combat scenarios where the combat target has entered the attack coverage area. It maximizes the immediacy of user commands and the continuity of attacks, avoids disrupting the combat rhythm due to unnecessary movement, and significantly improves the feedback of operations and the certainty of attacks.

[0222] Optionally, if the minimum distance between the torso at its initial position in the world coordinate system and the target is within the second attack range, it is determined that although the target is outside the direct attack range, it is still within the human-machine collaborative compensation range where a limited displacement can achieve an effective hit. The optimal movement path from the initial position to the second target position is calculated, ensuring that after the legged robot moves, its position allows its initial spatial position to fall precisely within the first attack range. Subsequently, the legged robot is controlled to complete this short-range displacement, and after the posture stabilizes, a pure range attack is initiated at the second target position. For example, when the distance is 1.8 meters and the effective radius of the sweep kick is 1.2 meters, it is calculated that the robot needs to move forward 0.6 meters to bring the target into an effective range within 1.5 meters before executing the sweep kick. This mechanism transforms an "infeasible attack" into an "executable attack" through minimal-cost spatial compensation, avoiding energy waste and trajectory loss due to blind pursuit, as well as the frustrating experience of "being so close yet unable to hit," enhancing intelligent adaptability and tactical flexibility under edge conditions.

[0223] Optionally, if the torso at its initial position in the world coordinate system is within the third attack range of the combatant, the combatant is determined to have exceeded the reasonable boundary of compensable movement. Any displacement will not achieve an effective hit before the combatant leaves the attack line of sight, and the movement process will significantly delay the attack response, disrupting the continuity of the attack intent. Therefore, the displacement strategy is abandoned, and a pure range attack action is executed directly at the initial position, allowing the swing trajectory to cover the area where the combatant may exist in the form of a wide-area release in place. For example, when the distance exceeds 2.5 meters, the combatant is determined to be within the third attack range. Even if a hit cannot be guaranteed, a range sweep kick or a 360° rotation attack is still executed to create a deterrent attack effect. The core of this strategy is not to pursue a hit, but to maintain the integrity of the user's attack intent expression, preventing the interaction gap and sense of loss of control caused by "no action". This mode provides a minimum effective response with "feedback and performance" for long-range combat. It is a proactive compromise and technical backup for user experience and tactical expression under physical constraints, ensuring that the combat process is continuous, stable, and predictable.

[0224] Figure 9 This is a schematic diagram of another legged robot according to an embodiment of this application, such as... Figure 9As shown, the legged robot 90 includes a torso 91; a plurality of target fighting limbs 92 and corresponding plurality of fighting limb ends 93 respectively connected to the torso; at least one visual perception module 94 configured to acquire visual information of the surrounding environment of the legged robot, the visual information of the surrounding environment including at least one fighting object; at least one communication module 95 configured to receive movement commands or fighting commands sent by a remote control device; at least one processor 96; and at least one memory 97, the at least one memory encoding instructions, which, when executed by at least one processor, cause the legged robot to: step S902, move in the target direction indicated by the received movement command sent by the remote control device; step S904, in response to the received fighting command sent by the remote control device, acquire the initial orientation of the forward face of the torso, and determine the starting orientation of the fighting action indicated by the fighting command to be executed; adjust the orientation of the forward face of the torso from the initial orientation to the starting orientation; and execute the fighting action in the fighting direction indicated by the fighting command in response to the orientation of the forward face of the torso being the starting orientation.

[0225] Regarding the aforementioned hardware, and Figure 1 The functions are similar, so they will not be described in detail here.

[0226] At least one memory is encoded with instructions that, when executed by at least one processor, cause the legged robot to: step S902, move in the target direction indicated by the received movement instruction sent by the remote control device; step S904, obtain the initial orientation of the torso in the world coordinate system and determine the starting orientation of the fighting action indicated by the fighting instruction to be executed in response to the received fighting instruction sent by the remote control device; adjust the orientation of the torso's forward face from the initial orientation to the starting orientation; and execute the fighting action in the fighting direction indicated by the fighting instruction in response to the orientation of the torso's forward face being the starting orientation, wherein the torso's forward face is the torso's forward face.

[0227] The aforementioned initial orientation refers to the orientation that the legged robot's torso needs to be adjusted to match the initial posture required by the combat action before executing a certain combat command. This initial orientation can be determined by the directional semantics carried by the combat command itself, independent of the real-time position or visual information of the combat opponent. Instead, it is pre-defined in the semantic meaning of the command to ensure that the combat action has the correct starting angle in its physical structure. For example, when the combat command is "right front straight punch," the initial orientation is a fixed position where the torso faces the right front of the combat opponent at a 45-degree angle. The aforementioned initial orientation can be the same as or different from the initial orientation.

[0228] The aforementioned fighting direction refers to the spatial orientation that the target limb's end effector should face when the legged robot performs an attack action, as explicitly specified by the fighting command. It defines the overall extension path of the swing trajectory and works in conjunction with the initial orientation of the torso to jointly determine the spatial orientation of the attack action. The fighting direction is independent of the position of the fighting target; it is determined by the content of the fighting command, such as "forward straight strike," "right sweep," or "upward chop." Based on this fighting direction, the principal axis vector of the swing trajectory is constructed in the world coordinate system, driving the fighting limb's end effector to complete the predetermined trajectory along that direction. The fighting direction is not updated with target movement and does not rely on visual feedback correction; it is essentially an inherent attribute of the fighting command, ensuring that the legged robot can still execute standardized attacks according to the user's intentions even without the perceptual constraint of a fighting target.

[0229] In this embodiment, when the legged robot executes combat commands sent by the remote control device, it constructs a combat execution mechanism that does not rely on the positional perception of the combat object but is based on the semantic drive of the combat commands. This mechanism, by clearly distinguishing the control logic between "movement commands" and "combat commands," enables the legged robot to accurately respond to the user's explicit commands for attack direction and starting posture even without visual target tracking, achieving efficient, stable, and predictable combat behavior. The core of this mechanism lies in shifting the basis of combat action execution from "environmental perception" to "command presets," thereby reducing the computational complexity of the legged robot and improving response speed and combat consistency.

[0230] Optionally, if the legged robot receives a combat command from a remote control device, it does not need to rely on the visual perception module to obtain the position of the target. Instead, it directly extracts the built-in combat direction from the combat command and combines it with the predefined motion model associated with the combat command to determine the necessary starting orientation for the combat action to be performed. The starting orientation is the torso orientation parameter pre-bound to the combat command in the protocol layer. For example, the starting orientation associated with the "right front straight punch" command is a torso facing 45 degrees to the right front of the robot, and the starting orientation associated with the "back spinning kick" command is a torso facing 180 degrees backward. The above starting orientation is fixed by the command encoding and is independent of the position of the target. It is set based on the kinematic feasibility of the action and human combat habits, ensuring that the legged robot starts in a preset posture regardless of the target's position. The embodiments of this application decouple the attack action from the environmental perception through the above mechanism, enabling the legged robot to stably execute the attack according to the user's intention in complex, occluded or dynamically unpredictable environments, avoiding action failure caused by visual misjudgment or loss of the fighting object, and significantly improving the robustness and reliability of the legged robot under non-ideal conditions.

[0231] Optionally, after determining the initial orientation, the orientation adjustment process of the forward face is initiated. During the orientation adjustment process, the spatial angle of the initial orientation of the torso can be obtained by aligning with the world coordinate system through the inertial measurement unit; subsequently, the angular deviation between the initial orientation and the starting orientation is calculated, and the torso rotation mechanism is driven to complete the posture correction along the optimal path until the angular error between the orientation of the forward face and the starting orientation is less than a preset tolerance threshold. This process completes closed-loop control based on the initial orientation defined by the combat command and the data from the self-attitude sensor. After confirming that the orientation of the forward face has been accurately aligned with the initial orientation, the execution of the combat action is initiated. According to the combat direction specified in the combat command, the main axis of the swing trajectory of the target combat limb is constructed in the world coordinate system. For example, if the combat direction is "60 degrees to the right front", the target combat limb is controlled to execute a preset fixed trajectory along the above combat direction, and the corresponding combat action is executed regardless of whether the combat opponent is located on the above fixed trajectory path. The above-mentioned method of executing combat actions ensures that the direction and form of the combat actions in space follow the user's needs, achieving deterministic control of "command is executed, direction is output", avoiding delays, jitters or false triggers caused by reliance on visual guidance, and ensuring the reproducibility of action execution and tactical consistency.

[0232] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0233] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0234] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0235] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0236] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0237] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0238] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A legged robot, characterized in that, include: trunk; Multiple target fighting limbs and corresponding multiple fighting limb ends connected to the torso; At least one visual perception module is configured to acquire visual information of the environment surrounding the legged robot, the visual information of the environment including at least one fighting object; At least one communication module is configured to receive movement commands and combat commands sent by the remote control device; At least one processor; as well as, At least one memory, the at least one memory being encoded with instructions that, when executed by the at least one processor, cause the legged robot to: In response to the received movement command sent by the remote control device, move in the target direction indicated by the movement command; The received fighting command sent by the remote control device is a first fighting command. The first fighting command includes target fighting action information. After obtaining the initial spatial position of the fighting object, the hit rate of multiple attack directions is evaluated through pre-calculation. If, from the multiple attack orientations, a first target orientation is selected that makes the probability of the attack hitting the fighting object in the initial spatial position of the legged robot greater than or equal to a probability threshold, then in the initial position, the orientation of the forward face of the torso is adjusted from the initial orientation to the first target orientation. During the process of adjusting the orientation of the front face of the torso from the initial orientation to the first target orientation, the current spatial position of the fighting object is continuously identified, and a first updated target orientation is continuously selected such that the probability of the attack hitting the fighting object in the current spatial position is greater than or equal to the probability threshold. In response to the orientation of the front face of the torso being the first updated target orientation, the attack action indicated by the first fighting command is executed in the direction of the current spatial position. Wherein, the initial spatial position is used to represent the spatial position of multiple points on the surface of the geometric bounding body of the fighting object in the world coordinate system when the fighting command sent by the remote control device is received, the current spatial position is used to represent the current position of multiple points on the surface of the geometric bounding body of the fighting object in the world coordinate system, and the attack action includes attack actions other than pure area attack actions.

2. The legged robot according to claim 1, characterized in that, The at least one memory is also encoded with additional instructions, which, when executed by the at least one processor, further cause the legged robot to: adjust the orientation of the forward face of the torso from an initial orientation to the first target orientation, including: If the first updated target orientation is an orientation that occurs during the process of adjusting from the initial orientation to the first target orientation, then the orientation of the front face of the torso is adjusted from the initial orientation to the first updated target orientation. In response to the orientation of the front face of the torso being the first updated target orientation, the attack action is performed in the direction of the current spatial position.

3. The legged robot according to claim 2, characterized in that, The at least one memory is further encoded with additional instructions, which, when executed by the at least one processor, cause the legged robot, in response to the orientation of the forward face of the torso being the first target orientation, to perform the attack action indicated by the first combat command in the direction of the current spatial position, including: If, during the process of adjusting the orientation of the torso's forward face to the first target orientation, the probability of the attack hitting the combatant at the current spatial position is less than the probability threshold, and the first updated target orientation is the orientation after adjusting from the initial orientation to the first target orientation, then the orientation of the torso's forward face is adjusted from the initial orientation to the first updated target orientation. In response to the orientation of the torso's forward face being the first updated target orientation, the attack is performed in the direction of the current spatial position.

4. The legged robot according to claim 3, characterized in that, The at least one memory is also encoded with additional instructions, which, when executed by the at least one processor, cause the legged robot, in response to the orientation of the forward face of the torso being the first target orientation, to perform an attack action indicated by the first combat command in the direction of the current spatial position, including: If, during the process of adjusting the orientation of the torso's forward face to the first target orientation, the probability of the attack hitting the fighting object at the current spatial position is less than the probability threshold, and no other orientation is detected, and the probability of the attack hitting the fighting object at the current spatial position is greater than or equal to the probability threshold, then after the orientation of the torso's forward face is adjusted to the first target orientation, the attack is performed in the direction of the current spatial position.

5. The legged robot according to claim 1, characterized in that, The at least one memory is also encoded with additional instructions, which, when executed by the at least one processor, cause the legged robot to obtain the initial spatial position of the fighting object in the world coordinate system from the visual information by performing the following steps: In response to the received first fighting instruction including information about the fighting object, a fighting object that satisfies the information of the fighting object is selected from at least one fighting object in the acquired visual information, and the initial spatial position of the fighting object that satisfies the information of the fighting object is obtained from the visual information; Before receiving the first fighting instruction, an instruction to determine the fighting target is received, and the initial spatial position of the fighting target indicated by the instruction of the fighting target is obtained from the acquired visual information; In response to the received first fighting instruction including fighting direction information, the fighting object is determined from the acquired visual information according to the fighting direction pointed to by the fighting direction information, and the initial spatial position of the fighting object in the fighting direction pointed to is obtained.

6. The legged robot according to claim 5, characterized in that, The at least one memory also encodes additional instructions, which, when executed by the at least one processor, further enable the legged robot to: After pre-calculating and evaluating the hit rate for multiple attack directions, it also includes: If, from the multiple attack orientations, no first target orientation is selected that, from the initial position of the legged robot, the probability of the attack hitting the fighting object at the initial spatial position is greater than or equal to the probability threshold, then the receiving of the movement command triggered by the remote control device is stopped, or the execution of the received movement command triggered by the remote control device is stopped, and the robot moves from the initial position to the first target position, and at the first target position, the hit rate of the multiple attack orientations is evaluated through pre-calculation; If, from the multiple attack orientations, a first target orientation is selected at the first target position of the legged robot, the probability of the attack hitting the fighting object at the initial spatial position is greater than or equal to the probability threshold, then at the first target position, the orientation of the forward face of the torso is adjusted from the initial orientation to the first target orientation, and in response to the orientation of the forward face of the torso being the first target orientation, the attack action is performed in the direction of the initial spatial position.

7. The legged robot according to claim 1, characterized in that, The combat target includes a head and a torso. The at least one memory also encodes additional instructions, which, when executed by the at least one processor, cause the legged robot to: pre-calculate and evaluate the hit rate for multiple attack orientations, including: If the attack action can hit the head or torso of the fighting object, the probability that the attack action hits the fighting object is greater than or equal to the probability threshold. If the attack action fails to hit the head and torso of the fighting target, it is determined that the probability of the attack action hitting the fighting target is less than the probability threshold.

8. The legged robot according to claim 1, characterized in that, The combat target includes a head, a torso, and multiple limbs connected to the torso. The at least one memory also encodes additional instructions, which, when executed by the at least one processor, cause the legged robot to: pre-calculate and evaluate the hit rate for multiple attack orientations, including: If the attack action can hit the head, torso, or at least one limb of the fighting object, the probability that the attack action hits the fighting object is greater than or equal to the probability threshold. If the attack action fails to hit the head, torso, or any of the limbs of the fighting target, the probability that the attack action hits the fighting target is less than the probability threshold.

9. The legged robot according to claim 1, characterized in that, The at least one memory also encodes additional instructions, which, when executed by the at least one processor, cause the legged robot to: perform a pre-calculated hit rate assessment for multiple attack orientations, including: If the minimum distance between the legged robot in the initial position and the fighting object is within the first distance range corresponding to the attack action to be performed, the process of performing the attack action under the multiple attack orientations is pre-calculated based on the initial position and the initial spatial position to obtain a hit rate evaluation result, wherein the hit rate evaluation result is used to represent the probability that the attack action hits the fighting object in the initial spatial position under the multiple attack orientations; If the minimum distance between the legged robot in the initial position and the fighting object is within the second distance range corresponding to the attack action to be performed, the attack action is performed in the initial position, wherein the minimum value of the second distance range is greater than the maximum value of the first distance range.

10. The legged robot according to claim 1, characterized in that, The at least one memory also encodes additional instructions, which, when executed by the at least one processor, further enable the legged robot to: Upon receiving a combat action stop command from the remote control device, the current offensive action is stopped and a defensive action is switched, wherein the current offensive action is either the attack action or the pure area attack action.

11. The legged robot according to claim 1, characterized in that, The at least one memory also encodes additional instructions, which, when executed by the at least one processor, further enable the legged robot to: During the execution of the fighting action indicated by the current fighting command sent by the remote control device, in response to receiving a cancel fighting command sent by the remote control device, and within the time range after triggering the cancel fighting command, receiving the next fighting command sent by the remote control device, the execution of the fighting action indicated by the current fighting command is stopped, and the fighting action indicated by the next fighting command is executed. or, During the execution of the fighting action indicated by the current fighting command sent by the remote control device, in response to receiving the next fighting command sent by the remote control device, after hitting the fighting target, the fighting action indicated by the next fighting command is executed. or, During the execution of the fighting action indicated by the current fighting command sent by the remote control device, in response to receiving the next fighting command sent by the remote control device, after completing the unfolding segment of the swing trajectory indicated by the current fighting command, the fighting action indicated by the next fighting command is executed.

12. The legged robot according to claim 1, characterized in that, The at least one memory also encodes additional instructions, which, when executed by the at least one processor, further enable the legged robot to: In response to receiving a set of fighting instructions sent by the remote control device, the multiple fighting instructions to be executed are parsed from the set of fighting instructions; According to the time sequence of receiving multiple combat commands, the combat actions indicated by the combat commands are executed sequentially in the direction of the current spatial location.

13. The legged robot according to claim 12, characterized in that, The additional instructions, when executed by the at least one processor, also enable the legged robot to: During the execution of the fighting action indicated by the current fighting command in the fighting command set, in response to hitting the fighting target, the fighting action indicated by the next fighting command of the current fighting command is executed in the direction of the current spatial position according to the time sequence. or, During the execution of the fighting action indicated by the current fighting command in the fighting command set, in response to completing the unfolding segment of the swing trajectory indicated by the current fighting command, the fighting action indicated by the next fighting command of the current fighting command is executed in the direction of the current spatial position according to the time sequence.

14. The legged robot according to claim 1, characterized in that, The combat limb extremities include at least one of the following: In response to the legged robot holding a fighting device at the end of its limb, the end of the fighting limb being the attack end of the fighting device, wherein the fighting device includes a first type of fighting device and a second type of fighting device, the attack end of the first type of fighting device being the attack end of the first type of fighting device, and the attack end of the second type of fighting device being the second type of fighting device itself. In response to the limb end not holding the fighting device, the fighting limb end is the limb end.

15. The legged robot according to claim 1, characterized in that, The types of combat commands include at least one of the following: command types triggered by physical buttons of the remote control device, command types triggered by virtual buttons of the remote control device, voice commands, and motion capture commands triggered by motion capture devices.

16. A legged robot, characterized in that, include: trunk; Multiple target fighting limbs and corresponding multiple fighting limb ends connected to the torso; At least one visual perception module is configured to acquire visual information of the environment surrounding the legged robot, the visual information of the environment including at least one fighting object; At least one communication module is configured to receive movement commands and combat commands sent by the remote control device; At least one processor; as well as, At least one memory, the at least one memory being encoded with instructions that, when executed by the at least one processor, cause the legged robot to: In response to the received movement command sent by the remote control device, move in the target direction indicated by the movement command; In response to the received combat command sent by the remote control device, the initial orientation of the forward face of the torso is obtained, and the starting orientation of the combat action indicated by the combat command to be executed is determined; Adjust the orientation of the forward face of the torso from the initial orientation to the starting orientation; In response to the orientation of the forward face of the torso as the initial orientation, the fighting action is executed in the fighting direction indicated by the fighting command; The step of determining the starting orientation of the fighting action indicated by the fighting command to be executed includes: extracting the built-in fighting direction from the fighting command, and combining it with the predefined motion model associated with the fighting command to determine the starting orientation of the fighting action to be executed. The fighting direction refers to the spatial direction that the end of the fighting limb should face when the legged robot performs the fighting action, as specified by the fighting command. It is used to define the overall extension path of the swing trajectory. The fighting direction is not updated with the movement of the fighting object, nor does it rely on visual feedback correction. It is essentially an inherent attribute of the fighting command.

Images