Robot control method, robot, and computer-readable storage medium

Abstract

The application is suitable for the technical field of robot control, and provides a robot control method, a robot and a computer readable storage medium, which comprises the following steps: obtaining a first instruction sequence of a robot corresponding to a current control time; calculating a safety confidence corresponding to the first instruction sequence according to preset first characteristic information; wherein the first characteristic information is used for representing a critical state of dangerous behavior of the robot; the safety confidence is used for representing a safety degree of robot behavior caused by the first instruction sequence; performing filtering processing on the first instruction sequence according to the safety confidence to obtain a second instruction sequence; and controlling the robot according to the second instruction sequence. Through the above method, the control accuracy and the operation safety of the robot can be effectively considered.

Description

Technical Field

[0001] This application belongs to the field of robot control technology, and in particular relates to a robot control method, a robot, and a computer-readable storage medium. Background Technology

[0002] Embodied intelligence is artificial intelligence with a physical entity. It relies on "body and multiple senses" to continuously interact with the real physical environment, forming a closed loop of perception, decision-making, action, environmental feedback, and relearning. It can autonomously complete physical operations and adapt to changing scenarios in the real world. The physical entity of embodied intelligence can be called an embodied intelligent robot. It is equipped with "sensory" devices such as cameras, tactile sensors, force sensors, and microphones, and performs actions through its external hardware (such as torso, arms, legs, hands, and joints).

[0003] Currently, in the process of human-computer interaction, embodied intelligent robots mainly rely on large models for decision-making. However, large models are prone to "big model illusion," which means generating non-existent environments, incorrect action logic, planning action instructions that cannot be implemented, or even planning action instructions that will cause the robot to collide, fall, or make operational errors. This leads to significant safety uncertainty risks in the practical application of embodied intelligent robots. Summary of the Invention

[0004] This application provides a robot control method, a robot, and a computer-readable storage medium, which can effectively balance the robot's control accuracy and operational safety.

[0005] In a first aspect, embodiments of this application provide a robot control method, including: Obtain the robot's instruction sequence corresponding to the current control moment to obtain the first instruction sequence; The safety confidence level corresponding to the first instruction sequence is calculated based on the preset first feature information; wherein, the first feature information is used to characterize the critical state of the robot's dangerous behavior; and the safety confidence level is used to characterize the safety level of the robot behavior caused by the first instruction sequence. The first instruction sequence is filtered according to the security confidence level to obtain the second instruction sequence; The robot is controlled according to the second instruction sequence.

[0006] In this embodiment, the risk level of the original first instruction sequence is quantified by extracting the first feature information corresponding to the critical state of the robot's dangerous behavior, thus obtaining a safety confidence level. Then, based on this safety confidence level, the original first instruction sequence is filtered to obtain a relatively safe second instruction sequence, which is used to control the robot. This method accurately quantifies motion risks and adjusts instructions as needed, ensuring robot safety while preserving the original operational intent to the greatest extent possible, thereby balancing robot control accuracy and operational safety.

[0007] In one possible implementation of the first aspect, calculating the security confidence level corresponding to the first instruction sequence based on preset first feature information includes: Obtain the state data corresponding to the current control moment; wherein, the state data includes the state information of the environment in which the robot is located; The second feature information is obtained by fusing and encoding the first instruction sequence and the state data; The security confidence level is calculated based on the first similarity between the first feature information and the second feature information.

[0008] In the above implementation method, the safety confidence level is calculated based on the second feature information that integrates the robot's own state and the environmental state, which fully combines multi-dimensional actual working conditions and greatly improves the authenticity and accuracy of the safety confidence level assessment.

[0009] In one possible implementation of the first aspect, calculating the security confidence based on the first similarity between the first feature information and the second feature information includes: Calculate the first similarity between the first feature information and the second feature information; Obtain the second similarity corresponding to each historical moment within the historical time window prior to the current moment; wherein, the second similarity corresponding to the historical moment is the similarity between the first feature information and the third feature information, and the third feature information is feature information generated based on the robot's instruction sequence and state data corresponding to the historical moment; The first similarity and the second similarity corresponding to each historical moment are weighted and summed according to the preset weights corresponding to each historical moment to obtain the third similarity; wherein, the preset weights are used to characterize the degree of time decay; The security confidence level is determined based on the third similarity.

[0010] The above implementation method is equivalent to introducing the similarity corresponding to historical moments and weighting the fusion according to the time decay weight. This can reduce the interference of instantaneous data fluctuations, make the security risk assessment have temporal continuity, and thus make the calculated security confidence more stable and reliable.

[0011] In one possible implementation of the first aspect, determining the security confidence level based on the third similarity includes: The third similarity is nonlinearly mapped according to preset parameters to obtain the security confidence score; The preset parameters include a first parameter and a second parameter. The first parameter is used to characterize the sensitivity to dangerous behavior, and the second parameter is used to characterize the steepness of the critical state from safety to danger.

[0012] In the above implementation method, the safety confidence level is obtained by completing the nonlinear similarity mapping with the help of two preset parameters. The sensitivity of dangerous behavior perception and the steepness of the safety hazard threshold switching can be flexibly adjusted to accurately adapt to the robot safety judgment requirements in different scenarios.

[0013] In one possible implementation of the first aspect, the step of filtering the first instruction sequence according to the security confidence level to obtain the second instruction sequence includes: The security status level corresponding to the first instruction sequence is determined based on the security confidence level. The first instruction sequence is filtered according to the filtering strategy corresponding to the security status level to obtain the second instruction sequence.

[0014] In the above implementation method, the safety level is divided according to the safety confidence level, and the corresponding filtering strategy is matched to process the original first instruction sequence. The robot operation instructions can be hierarchically controlled, taking into account both operation safety and operation smoothness.

[0015] In one possible implementation of the first aspect, the step of filtering the first instruction sequence according to the filtering strategy corresponding to the security state level to obtain the second instruction sequence includes: If the security status level is the first level, then the first instruction sequence is recorded as the second instruction sequence; If the security status level is the second level, then the first instruction sequence is subjected to overload constraints to generate the second instruction sequence; If the security status level is level three, then the second instruction sequence is set to empty; The security confidence level corresponding to the first level is higher than that corresponding to the second level, and the security confidence level corresponding to the second level is higher than that corresponding to the third level.

[0016] The above implementation method is divided into three levels of processing instructions: no correction, overload constraint, and direct clear. Low risk means no intervention, medium risk means gentle speed and force limiting, and high risk means direct shutdown, balancing the smoothness of operation and the safety of extreme working conditions.

[0017] In one possible implementation of the first aspect, the overload constraint on the first instruction sequence to generate the second instruction sequence includes: The first instruction sequence is proportionally adjusted based on the security confidence level to obtain the adjustment instruction; The adjustment instructions are optimized using preset rules as constraints to obtain the second instruction sequence; wherein the preset rules are used to describe the safety boundary conditions during the robot's movement.

[0018] The above implementation involves two levels of instruction constraint optimization. First, the first instruction sequence is adjusted proportionally based on the safety confidence level. This allows for adaptive adjustment of the action amplitude according to the level of safety risk, reducing the conservative action or risk omission caused by fixed amplitude limits. Second, the optimized instructions are constrained by preset rules that include various motion safety boundary conditions of the robot. This ensures that the final output second instruction sequence can dynamically adapt to the motion level according to safety risks, while strictly adhering to safety red lines such as hardware limits, stability limits, and environmental restricted areas. This significantly improves the safety redundancy of robot operations, preventing instability, collisions, joint overloads, and other malfunctions, while avoiding indiscriminate restriction of movement. It maximizes the retention of the robot's normal operational flexibility and efficiency, achieving a two-way balance between safety protection and operational performance.

[0019] In one possible implementation of the first aspect, the method further includes: Acquire scene information for various hazardous scenarios; wherein, the scene information for each hazardous scenario includes a parameter dictionary and a topology map of the hazardous scenario; the parameter dictionary of the hazardous scenario includes text information describing the robot's behavior in the hazardous scenario and corresponding physical control parameters; the topology map of the hazardous scenario includes the spatiotemporal relationship between the robot and other objects in the environment in the hazardous scenario; Obtain preset physical boundary information; wherein, the physical boundary information is used to describe the physical boundary conditions during the robot's movement. The first feature information is generated based on the physical boundary information and the scene information of each of the various dangerous scenarios.

[0020] The above implementation method collects hazardous scene information from multiple dimensions, relies on a parameter dictionary to clarify the robot's action textual definitions and hardware control parameters under various risk scenarios, and uses a topology map to clearly reconstruct the temporal and spatial positional relationships between the robot and surrounding objects. Simultaneously, it combines this with the robot's movement-specific physical boundary information, fusing the two types of data to generate the first feature information. This approach achieves a complete binding of hazardous scene behavioral parameters, spatiotemporal positional relationships, and hardware movement limitations, making risk no longer a vague concept but fully quantified and concrete. This facilitates subsequent accurate assessment of risk levels and matching of safety constraint rules, comprehensively covering various safety hazards such as joint overload, collisions, and instability. It avoids misjudgments and omissions caused by missing information in a single dimension, providing comprehensive, quantifiable, and traceable data for subsequent instruction grading and limiting, and safety optimization. This ensures that robot safety management has complete scenario support, improving the accuracy, completeness, and feasibility of safety control.

[0021] Secondly, embodiments of this application provide a robot control device, including: The acquisition unit is used to acquire the robot's instruction sequence corresponding to the current control moment, and obtain the first instruction sequence; A calculation unit is configured to calculate the safety confidence level corresponding to the first instruction sequence based on preset first feature information; wherein, the first feature information is used to characterize the critical state of the robot's dangerous behavior; and the safety confidence level is used to characterize the safety level of the robot's behavior caused by the first instruction sequence. A filtering unit is used to filter the first instruction sequence according to the security confidence level to obtain a second instruction sequence; A control unit for controlling the robot according to the second instruction sequence.

[0022] Thirdly, embodiments of this application provide a robot, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the robot control method as described in any one of the first aspects above.

[0023] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the robot control method as described in any one of the first aspects above.

[0024] Fifthly, embodiments of this application provide a computer program product that, when run on a terminal device, causes the terminal device to execute the robot control method described in any one of the first aspects.

[0025] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart illustrating the robot control method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the topology of a hazardous scene provided in an embodiment of this application; Figure 3 This is a schematic diagram of the calculation process for security confidence provided in the embodiments of this application; Figure 4 This is a structural block diagram of the robot control device provided in the embodiments of this application; Figure 5 This is a schematic diagram of the robot provided in the embodiments of this application. Detailed Implementation

[0028] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0029] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0030] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0031] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0032] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0033] References to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0034] Embodied intelligence is artificial intelligence with a physical entity. It relies on its body and multiple senses to continuously interact with the real physical environment, forming a closed loop of perception, decision-making, action, environmental feedback, and relearning. It autonomously performs physical operations and adapts to changing scenarios in the real world. The physical entity of embodied intelligence can be called an embodied intelligent robot. It is equipped with sensory devices such as cameras, tactile sensors, force sensors, and microphones, and performs actions through its physical hardware (such as a torso, arms, legs, hands, and joints). Common embodied intelligent robots include humanoid robots, robotic arms, quadrupedal robotic dogs, and surgical robots.

[0035] Currently, embodied intelligent robots primarily rely on large models for decision-making during human-computer interaction. For example, they use large language models (LLMs) or multimodal visual language models (VLMs) to determine the robot's action commands. However, large models are prone to "big model illusion," which generates non-existent environments, incorrect action logic, and plans action commands that cannot be implemented, or even commands that could cause the robot to collide, fall, or malfunction. This leads to significant safety uncertainties in the practical application of embodied intelligent robots. For instance, faced with a rusty bolt, a large model might output highly rigid motion control commands with a tendency towards violent destruction. Furthermore, due to the distortion of the semantics of the visual environment, a large model might output unintended action commands that could cause physical harm to nearby human operators.

[0036] In related technologies, the safety control of embodied intelligent robots mainly relies on underlying hard mechanical collision avoidance or simple electronic fences. This method is a post-event or end-point protection and cannot dynamically intercept and flexibly correct uncertain action commands in a timely manner at the intention planning level. As a result, embodied intelligent robots face huge safety uncertainty risks when they are actually put into production.

[0037] Based on this, this application provides a robot control method. In this embodiment, the risk level of the original first instruction sequence is quantified by extracting first feature information corresponding to the critical state of the robot's dangerous behavior, thus obtaining a safety confidence level. Then, based on this safety confidence level, the original first instruction sequence is filtered to obtain a relatively safe second instruction sequence, and the robot is controlled according to the second instruction sequence. Through this method, motion risks can be accurately quantified and instructions can be adjusted as needed, ensuring robot operational safety while preserving the original operational intent to the greatest extent possible, thereby balancing robot control accuracy and operational safety.

[0038] For ease of description, robots are used in the following embodiments.

[0039] See Figure 1 This is a flowchart illustrating the robot control method provided in an embodiment of this application. It is intended as an example and not a limitation. The method may include the following steps: S101, Obtain the robot's instruction sequence corresponding to the current control moment to obtain the first instruction sequence.

[0040] The first instruction sequence can be the instruction sequence output by the large model responsible for deciding the robot's action commands. Optionally, the large model can be an LLM or a VLM. For example, the large model can be deployed on the robot, and the robot's processor / controller inputs the sensor signals collected by the various sensors on the robot into the large model for decision-making to obtain the first instruction sequence. Alternatively, the large model can be deployed on a cloud server, and the robot interacts with the cloud server, sending the sensor signals collected by the robot's various sensors to the cloud server; the cloud server inputs the received sensor signals into the large model, outputs the first instruction sequence, and sends the first instruction sequence to the robot.

[0041] S102, calculate the security confidence level corresponding to the first instruction sequence based on the preset first feature information.

[0042] The first feature information is used to characterize the critical state of the robot's dangerous behavior. The safety confidence level is used to characterize the safety level of the robot's behavior caused by the first instruction sequence. It can be understood that the higher the safety confidence level, the higher the safety level of the robot's behavior caused by the first instruction sequence; the lower the safety confidence level, the lower the safety level of the robot's behavior caused by the first instruction sequence.

[0043] In some embodiments, the method for obtaining the first feature information includes: Obtain scene information for various hazardous scenarios; Obtain the preset physical boundary information; First feature information is generated based on physical boundary information and scene information of various hazardous scenarios.

[0044] The scenario information for each hazardous scenario includes a parameter dictionary and a topology map of the hazardous scenario; the parameter dictionary of the hazardous scenario includes text information describing the robot's behavior in the hazardous scenario and the corresponding physical control parameters; the topology map of the hazardous scenario includes the spatiotemporal relationship between the robot and other objects in the environment in the hazardous scenario.

[0045] Understandably, large models typically process unstructured natural language text, while robot controllers (such as physical controllers or servo drives) recognize specific physical values. In this embodiment, a parameter dictionary is used to bind textual intents, safety instructions, or hazard warnings describing robot behavior to underlying physical control parameters.

[0046] Optionally, the text information in the parameter dictionary may include semantic text identifiers, such as action descriptions, intent labels, and / or safety warnings. The physical control parameters in the parameter dictionary may include physical constraint types, physical boundary values, and units; where physical constraint types may include velocity, acceleration, torque, stiffness, damping, force thresholds, etc.; physical boundary values ​​and units are used to describe the exact upper and / or lower limits of various physical constraints. Additionally, the parameter dictionary may include effective conditions and context, used to describe the effective conditions of parameter pairs, wherein a set of parameter pairs includes a set of text information and its corresponding physical control parameters.

[0047] For example, a set of parameter pairs in the parameter dictionary used to describe dangerous and destructive actions is shown below.

[0048] Semantic text: "forcefully press down", "pry open the bolt with maximum torque", "slam into the workpiece at high speed".

[0049] Physical control parameters: Ultimate downward force (F_z_max): >200 N; Vertical velocity of the end effector (v_z_max): >1.2 m / s; Joint stiffness matrix coefficient (K_p): >5000 N / m (in an extremely high stiffness state); The aforementioned parameters have low security confidence levels. In other words, if the instruction sequence output by the large model reveals such parameter combinations, subsequent filtering processes can block or trigger rewriting.

[0050] For example, a set of parameter pairs in the parameter dictionary used to describe dangerous actions in human-machine collaborative scenarios is shown below.

[0051] Semantic text: "Human-machine collaborative safety mode", "Working close to human operators".

[0052] Physical control parameters: Maximum end-synthesis velocity (vTCP_max): <= 0.25 m / s; Instantaneous maximum impact force (F_contact_max): <= 140 N (human chest / arm safety tolerance limit); Joint soft limit reduction ratio (Range_scale): 80% (actively shrinks the workspace to prevent excessive arm swing).

[0053] The above parameters can be used for subsequent overload constraints.

[0054] It should be noted that if the first feature information only includes a parameter dictionary, the robot may not understand spatiotemporal relationships. For example, a high-speed arm swing might be safe in an open area but dangerous in a narrow passage. In this embodiment, a topology graph is used to describe the spatiotemporal relationships between the robot and other objects in the environment in a hazardous scenario, thereby improving the robot's ability to understand spatiotemporal relationships. Simply put, a topology graph can describe physical entities, safety boundaries, and spatiotemporal relationships as nodes and edges, used to describe what object is dangerous when it is near another object at what location and in what state.

[0055] Optionally, the topology graph may include entity nodes, node attributes, topological edge relationships, and a hazardous spatiotemporal envelope. Entity nodes may include the robot itself (e.g., each link joint), human operators (e.g., head, hands, torso), production line equipment (e.g., high-voltage cabinets, conveyor belts, jigs), and geofencing zones. Node attributes may include bounding box dimensions, 3D point cloud data, and motion state vectors (e.g., position, velocity, acceleration). Topological edge relationships describe the relative geometric relationships between nodes, such as relative distance vectors, inside / outside relationships, and relative approach velocities. The hazardous spatiotemporal envelope refers to a dynamically calculated safety buffer zone.

[0056] For example, see Figure 2 This is a schematic diagram of a topology of a hazardous scenario provided in an embodiment of this application. Figure 2 The diagram shown illustrates the topology of a human-machine collaborative fall / boundary crossing scenario in a confined workstation. Entity nodes are defined as follows: Node 1 (robot itself): Real-time position P_robot, current speed v_robot = 1.0 m / s; Node 2 (Human Operator): Real-time 3D skeleton point cloud (head and chest coordinates P_human); Node 3 (Fixed Equipment): Production Line High Voltage Control Cabinet; Node 4 (Restricted Area): Production Line Safety and Safety Restricted Area.

[0057] Topological edge relationships: Edge E_12 (Robots and Humans): Calculating Real-Time Distance Vectors And the relatively close velocity v_rel.

[0058] Edge E_13 (Robot and Fixed Equipment): Detects whether the robot trajectory prediction line intersects spatially with node 3.

[0059] The hazard assessment logic based on the above topology diagram can be as follows: If detected in the topology map If v_rel is less than 0.5 and v_rel > 0.5, a personal injury warning is triggered. If a spatial intersection is detected between the robot trajectory prediction line in the topology graph and node 3, a device collision warning is triggered.

[0060] In this embodiment, physical boundary information is used to describe the physical boundary conditions during robot movement. In short, it transforms the invisible and intangible safety red lines and physical constraints of the physical world into physical boundary conditions that computers can perform geometric calculations and vector encoding. For example, a large model can understand "don't bump into people," but the underlying physical controller can only understand three-axis coordinates, torque values, and volume space. Physical boundary conditions are parameters obtained by geometrically and dynamically quantifying abstract safety rules and the real-time perceived physical space.

[0061] Optionally, physical boundary conditions include geometric and spatial boundaries, dynamic and mechanical critical boundaries, and spatiotemporal topological and semantic barrier boundaries.

[0062] In this context, geometric and spatial boundaries describe the volumetric regions in three-dimensional space that the robot and its limbs must absolutely not intrude upon or must maintain a distance from. For example, geometric and spatial boundaries can include Oriented Bounding Boxes (OBBs), swept volume boundaries, and safety expansion bubbles. An Oriented Bounding Box is a three-dimensional cuboid boundary that tightly wraps around a human operator, surrounding precision equipment, or fixture using 3D point clouds. The bounding box contains the three-dimensional coordinates (x, y, z) of its center point, its length, width, and height dimensions (l, w, h), and its rotational orientation quaternions (q_x, q_y, q_z, q_w). A swept volume boundary is the dynamic volumetric envelope that the entire limb sweeps through in three-dimensional space when the robot arm executes a certain motion trajectory. A safety expansion bubble is a "buffer zone" that dynamically expands with speed outside the human or equipment OBB according to preset standards (e.g., when the robot's speed is 1 m / s, the boundary automatically expands outward by 0.3 m).

[0063] Dynamical and mechanical critical boundaries describe the physical limits that joints and end effectors must never exceed when a robot interacts with its environment. For example, dynamical and mechanical critical boundaries may include critical constants for end-effector contact force and pressure, joint limiting torque and angular acceleration, and the center of mass momentum envelope. The critical constants for end-effector contact force and pressure are defined based on the safety tolerance limits of different parts of the human body (e.g., 140 N for the chest, 110 N for the arm) and the upper limit of contact pressure in the three-dimensional direction of the end effector. Joint limiting torque and angular acceleration refer to the peak torque and maximum angular acceleration allowed by the servo motors of each joint of the robot. The center of mass momentum envelope refers to the boundary values ​​of the three-dimensional linear and angular momentum of the center of mass that ensure the humanoid robot does not fall when standing or walking on two legs.

[0064] Spatiotemporal topology and semantic barrier boundaries are used to describe logical forbidden zones within a specific time and environmental context. For example, spatiotemporal topology and semantic barrier boundaries can include 3D electronic fences and interactive spatiotemporal window boundaries. A 3D electronic fence can be an absolute forbidden voxel delineated on a map using an octomap, such as "the 3D space within 0.5m around a high-voltage distribution box." Interactive spatiotemporal window boundaries are predicted boundaries that incorporate the time dimension.

[0065] In some implementations, the step of generating the first feature information may include: The physical boundary information and the scene information of each of the various dangerous scenarios are encoded to obtain the feature vectors corresponding to each of the various dangerous scenarios; the feature vectors corresponding to each of the various dangerous scenarios are clustered to extract the first feature information.

[0066] Optionally, a multimodal model can be used to encode physical boundary information and scene information for each of the various hazardous scenarios. The multimodal model maps the physical boundary information and the scene information for each hazardous scenario to the same "hazard point" in a high-dimensional space, generating a fixed-dimensional feature vector for each hazardous scenario. In this way, during subsequent filtering, regardless of the generated instruction sequence, as long as it falls within this hazardous high-dimensional space after encoding, it can be promptly determined that "this action crosses a red line in the physical world," and an immediate interception can be triggered.

[0067] Optionally, the clustering process may include: first dividing multiple feature vectors into K subspaces based on the action type (e.g., "rapid arm swing," "high-pressure slamming," "boundary intrusion"); and then performing a density-based spatial clustering algorithm within each subspace. In this way, each cluster not only retains the cluster center point but also retains the edge sample points that are farthest from the cluster center and closest to the safety / danger threshold.

[0068] Optionally, after clustering, the first feature information can be extracted using a support vector machine (SVM) or convex hull algorithm. Specifically, the feature vectors corresponding to the clusters are used as positive samples, and similar "legal and safe action vectors" are introduced as negative samples. An SVM is trained in a high-dimensional space based on the positive and negative samples to find a decision hyperplane equation that accurately separates safety from danger. The first feature information is then extracted based on this decision hyperplane equation. Here, only support vectors that fall exactly on the margin boundary can be retained; these support vectors represent the most sensitive and dangerous critical action features.

[0069] The extracted support vectors are often highly correlated (i.e., the basis is not independent). If they are directly stored in the matrix, it will lead to a large number of repeated projection operations in subsequent calculations, wasting the chip's computing power.

[0070] To address the aforementioned issues, alternatively, after extracting the support vectors falling on the margin boundaries, redundancy removal can be performed. Specifically, singular value decomposition is performed on the matrix formed by the extracted support vectors to obtain the decomposition matrix; the decomposition matrix is ​​orthogonalized to obtain the orthogonal matrix; and the orthogonal matrix is ​​normalized to obtain the first feature information.

[0071] The aforementioned method for acquiring the first feature information involves collecting hazardous scene information from multiple dimensions. It relies on a parameter dictionary to clearly define the robot's actions and hardware control parameters under various risk scenarios. A topology map clearly reconstructs the temporal and spatial relationships between the robot and surrounding objects. Simultaneously, it incorporates the robot's movement-specific physical boundary information, fusing these two types of data to generate the first feature information. This approach achieves a complete binding of hazardous scene behavioral parameters, spatiotemporal positional relationships, and hardware movement limitations, making risk no longer a vague concept but fully quantifiable and concrete. This facilitates accurate subsequent risk level assessment and matching of safety constraint rules, comprehensively covering various safety hazards such as joint overload, collisions, and instability. It avoids misjudgments and omissions caused by missing information in a single dimension, providing comprehensive, quantifiable, and traceable data for subsequent instruction grading and limiting, and safety optimization. This ensures complete scenario support for robot safety management, improving the accuracy, completeness, and feasibility of safety control.

[0072] In some embodiments, see Figure 3 This is a schematic diagram illustrating the calculation process of security confidence level provided in an embodiment of this application. It is intended as an example and not a limitation. Figure 3 As shown, S102 includes: S201, Obtain the status data corresponding to the current control moment.

[0073] The state data includes the state information of the robot's environment. Optionally, the state data may include environmental point clouds. For example, an environmental point cloud is a collection of 3D coordinate points obtained by the robot's LiDAR and / or depth camera. Optionally, the state data may include the topological features of the environmental point cloud. For example, topological features include adjacency topology (used to describe whether different object regions in the point cloud are adjacent and have contact surfaces), connectivity topology (used to determine whether a point cloud is continuous), containment topology (used to determine whether the point cloud of one object is entirely within the space enclosed by another object), and occlusion topology (used to record the spatial hierarchy between two object point clouds), etc. It can be understood that the topological features of the environmental point cloud extract the spatial connection logic of the environment from the 3D point cloud, rather than simply the length, width, and height. Topological features do not focus on absolute geometric values ​​such as size, distance, length, etc., but only on the connectivity, containment, adjacency, occlusion, and belonging relationships between objects.

[0074] Optionally, the state data may also include the robot's state information. For example, the state data may include the robot's joint motion data, pose data, actuator hardware status (such as battery level, joint limits, etc.), and interaction status (such as ambient sound picked up by the microphone, WiFi signal quality, etc.).

[0075] S202, the second feature information is obtained by fusing and encoding the first instruction sequence and the status data.

[0076] Optionally, a lightweight model can be used to fuse and encode the first instruction sequence and state data to obtain a second feature information with fixed dimensions.

[0077] S203, calculate the security confidence level based on the first similarity between the first feature information and the second feature information.

[0078] In the above implementation method, the safety confidence level is calculated based on the second feature information that integrates the robot's own state and the environmental state, which fully combines multi-dimensional actual working conditions and greatly improves the authenticity and accuracy of the safety confidence level assessment.

[0079] In some implementations, S203 includes: Calculate the first similarity between the first feature information and the second feature information; Obtain the second similarity score for each historical moment within the historical time window preceding the current moment; The first similarity and the second similarity corresponding to each historical moment are weighted and summed according to the preset weights to obtain the third similarity. The security confidence level is determined based on the third similarity.

[0080] Among them, the second similarity corresponding to the historical moment is the similarity between the first feature information and the third feature information, and the third feature information is the feature information generated based on the robot's instruction sequence and state data corresponding to the historical moment.

[0081] The preset weights are used to characterize the degree of time decay.

[0082] Optionally, the cosine similarity between the first feature information and the second feature information can be calculated to obtain the first similarity. For example, according to the formula... Calculate the first similarity score. Wherein, The similarity between the second feature information and the i-th dangerous scene. This is the feature vector corresponding to the i-th dangerous scenario in the first feature information. This is the second feature information. It can be understood that the first similarity includes the similarity between the second feature information and each hazardous scenario in the first feature information.

[0083] Since the robot's actions are a series of temporal signals, single-frame comparisons are prone to false alarms. To address this issue, in this embodiment, a historical time window (e.g., 50ms, for a total of 50 frames of control signals) is maintained. The semantic changes in the temporal sequence are characterized by weighted summation of the second similarity at each historical moment within the historical time window.

[0084] For example, according to the formula Calculate the third similarity. Wherein, This is the third feature information. As a historical time window, Let be the second similarity corresponding to the i-th dangerous scenario at the t-th historical moment. This represents the preset weight corresponding to the t-th historical moment. The closer the historical moment is to the current moment, the higher the weight. The larger the value.

[0085] The above implementation method is equivalent to introducing the similarity corresponding to historical moments and weighting the fusion according to the time decay weight. This can reduce the interference of instantaneous data fluctuations, make the security risk assessment have temporal continuity, and thus make the calculated security confidence more stable and reliable.

[0086] In some implementations, the steps for determining the security confidence level based on the third similarity include: The third similarity is nonlinearly mapped according to preset parameters to obtain the security confidence level.

[0087] The preset parameters include a first parameter and a second parameter. The first parameter is used to characterize the sensitivity to dangerous behavior, and the second parameter is used to characterize the steepness of the critical state from safety to danger.

[0088] For example, according to the formula Perform nonlinear mapping. For safety confidence level, As the first parameter, This is the second parameter. It is understandable that... The smaller the value, the more sensitive the system is to danger; The larger the value, the steeper the transition from safety to danger.

[0089] In the above implementation method, the safety confidence level is obtained by completing the nonlinear similarity mapping with the help of two preset parameters. The sensitivity of dangerous behavior perception and the steepness of the safety hazard threshold switching can be flexibly adjusted to accurately adapt to the robot safety judgment requirements in different scenarios.

[0090] S103, the first instruction sequence is filtered according to the security confidence level to obtain the second instruction sequence.

[0091] In some embodiments, S103 includes: Determine the security status level corresponding to the first instruction sequence based on the security confidence level; The first instruction sequence is filtered according to the filtering strategy corresponding to the security status level to obtain the second instruction sequence.

[0092] In the above implementation method, the safety level is divided according to the safety confidence level, and the corresponding filtering strategy is matched to process the original first instruction sequence. The robot operation instructions can be hierarchically controlled, taking into account both operation safety and operation smoothness.

[0093] In some implementations, if the security confidence level is higher than a first threshold, the security status level is Level 1; if the security confidence level is higher than a second threshold but lower than the first threshold, the security status level is Level 2; and if the security confidence level is lower than the second threshold, the security status level is Level 3. The first threshold is greater than the second threshold. The security confidence level corresponding to Level 1 is higher than that corresponding to Level 2, and the security confidence level corresponding to Level 2 is higher than that corresponding to Level 3.

[0094] In some implementations, the filtering process based on the filtering strategy corresponding to the security state level includes three cases: I-III, as detailed below.

[0095] I. If the security status level is Level 1, then the first instruction sequence is recorded as the second instruction sequence.

[0096] II. If the security status level is Level 2, then overload constraints are applied to the first instruction sequence to generate the second instruction sequence.

[0097] III. If the security status level is level 3, then set the second instruction sequence to empty.

[0098] The above implementation method is divided into three levels of processing instructions: no correction, overload constraint, and direct clear. Low risk means no intervention, medium risk means gentle speed and force limiting, and high risk means direct shutdown, balancing the smoothness of operation and the safety of extreme working conditions.

[0099] In some implementations, II includes: The first instruction sequence is proportionally adjusted based on the safety confidence level to obtain the adjustment instruction; the adjustment instruction is optimized using preset rules as constraints to obtain the second instruction sequence; wherein, the preset rules are used to describe the safety boundary conditions during the robot's movement.

[0100] Optionally, the first instruction sequence includes speed. The step of proportionally adjusting the speed based on the security group confidence level may include: calculating a time scaling factor based on the security confidence level; and correcting the speed based on the time scaling factor. For example, according to the formula... Calculate the time scaling factor, where, The first threshold, These are preset parameters. According to the formula... Correction speed, among which, The speed in the first instruction sequence, The adjusted speed is the corrected speed; the adjustment command includes the corrected speed.

[0101] Optionally, the first instruction sequence includes stiffness. The step of proportionally adjusting the stiffness based on the safety group confidence level may include: according to the formula... ,in, The stiffness in the first instruction sequence, The adjustment command includes the corrected stiffness for the adjusted stiffness.

[0102] Optionally, the first instruction sequence includes damping. The step of proportionally adjusting the damping based on the safety group confidence level may include: according to the formula... ,in, For damping in the first instruction sequence. For the corrected damping, the adjustment command includes the corrected damping.

[0103] Optional, according to the formula The adjustment instructions were optimized. Among them, To adjust the instructions, This is the second instruction sequence. and These are parameters determined according to preset rules. It is understandable that... and It can be a matrix or a vector. and It is determined based on multiple safety boundary conditions in the preset rules.

[0104] Optionally, preset rules may include joint-level physical saturation limits, such as the absolute maximum permissible speed, maximum continuous output torque, and absolute spatial angle soft limits based on the mechanical limit jamming position of each degree of freedom (DoF) motor.

[0105] Optionally, preset rules may include the kinematic envelope of the center of mass based on collision-free foot support, such as an absolutely stable polygonal boundary calculated based on the real-time center of mass position and zero moment point.

[0106] Optionally, preset rules may include electronic fence restricted areas based on environmental entity point clouds, such as restricted spaces that no robot link may intrude into, defined by combining a local octree map updated in real time with onboard sensors (such as the human workstation space next to the production line).

[0107] The above implementation involves two levels of instruction constraint optimization. First, the first instruction sequence is adjusted proportionally based on the safety confidence level. This allows for adaptive adjustment of the action amplitude according to the level of safety risk, reducing the conservative action or risk omission caused by fixed amplitude limits. Second, the optimized instructions are constrained by preset rules that include various motion safety boundary conditions of the robot. This ensures that the final output second instruction sequence can dynamically adapt to the motion level according to safety risks, while strictly adhering to safety red lines such as hardware limits, stability limits, and environmental restricted areas. This significantly improves the safety redundancy of robot operations, preventing instability, collisions, joint overloads, and other malfunctions, while avoiding indiscriminate restriction of movement. It maximizes the retention of the robot's normal operational flexibility and efficiency, achieving a two-way balance between safety protection and operational performance.

[0108] S104, control the robot according to the second instruction sequence.

[0109] In this embodiment, the risk level of the original first instruction sequence is quantified by extracting the first feature information corresponding to the critical state of the robot's dangerous behavior, thus obtaining a safety confidence level. Then, based on this safety confidence level, the original first instruction sequence is filtered to obtain a relatively safe second instruction sequence, which is used to control the robot. This method accurately quantifies motion risks and adjusts instructions as needed, ensuring robot safety while preserving the original operational intent to the greatest extent possible, thereby balancing robot control accuracy and operational safety.

[0110] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0111] Corresponding to the robot control method described in the above embodiments, Figure 4 This is a structural block diagram of the robot control device provided in the embodiments of this application. For ease of explanation, only the parts related to the embodiments of this application are shown.

[0112] Reference Figure 4 The device 4 includes: The acquisition unit 41 is used to acquire the robot's instruction sequence corresponding to the current control moment, and obtain the first instruction sequence.

[0113] The calculation unit 42 is used to calculate the safety confidence level corresponding to the first instruction sequence based on preset first feature information; wherein, the first feature information is used to characterize the critical state of the robot's dangerous behavior; and the safety confidence level is used to characterize the safety level of the robot behavior caused by the first instruction sequence.

[0114] The filtering unit 43 is used to filter the first instruction sequence according to the security confidence level to obtain the second instruction sequence.

[0115] Control unit 44 is used to control the robot according to the second instruction sequence.

[0116] Optionally, the computing unit 42 is also used for: Obtain the state data corresponding to the current control moment; wherein, the state data includes the state information of the environment in which the robot is located; The second feature information is obtained by fusing and encoding the first instruction sequence and the state data; The security confidence level is calculated based on the first similarity between the first feature information and the second feature information.

[0117] Optionally, the computing unit 42 is also used for: Calculate the first similarity between the first feature information and the second feature information; Obtain the second similarity corresponding to each historical moment within the historical time window prior to the current moment; wherein, the second similarity corresponding to the historical moment is the similarity between the first feature information and the third feature information, and the third feature information is feature information generated based on the robot's instruction sequence and state data corresponding to the historical moment; The first similarity and the second similarity corresponding to each historical moment are weighted and summed according to the preset weights corresponding to each historical moment to obtain the third similarity; wherein, the preset weights are used to characterize the degree of time decay; The security confidence level is determined based on the third similarity.

[0118] Optionally, the computing unit 42 is also used for: The third similarity is nonlinearly mapped according to preset parameters to obtain the security confidence score; The preset parameters include a first parameter and a second parameter. The first parameter is used to characterize the sensitivity to dangerous behavior, and the second parameter is used to characterize the steepness of the critical state from safety to danger.

[0119] Optionally, filter unit 43 is also used for: The security status level corresponding to the first instruction sequence is determined based on the security confidence level. The first instruction sequence is filtered according to the filtering strategy corresponding to the security status level to obtain the second instruction sequence.

[0120] Optionally, filter unit 43 is also used for: If the security status level is the first level, then the first instruction sequence is recorded as the second instruction sequence; If the security status level is the second level, then the first instruction sequence is subjected to overload constraints to generate the second instruction sequence; If the security status level is level three, then the second instruction sequence is set to empty; The security confidence level corresponding to the first level is higher than that corresponding to the second level, and the security confidence level corresponding to the second level is higher than that corresponding to the third level.

[0121] Optionally, filter unit 43 is also used for: The first instruction sequence is proportionally adjusted based on the security confidence level to obtain the adjustment instruction; The adjustment instructions are optimized using preset rules as constraints to obtain the second instruction sequence; wherein the preset rules are used to describe the safety boundary conditions during the robot's movement.

[0122] Optionally, the computing unit 42 is also used for: Acquire scene information for various hazardous scenarios; wherein, the scene information for each hazardous scenario includes a parameter dictionary and a topology map of the hazardous scenario; the parameter dictionary of the hazardous scenario includes text information describing the robot's behavior in the hazardous scenario and corresponding physical control parameters; the topology map of the hazardous scenario includes the spatiotemporal relationship between the robot and other objects in the environment in the hazardous scenario; Obtain preset physical boundary information; wherein, the physical boundary information is used to describe the physical boundary conditions during the robot's movement. The first feature information is generated based on the physical boundary information and the scene information of each of the various dangerous scenarios.

[0123] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0124] in addition, Figure 4 The robot control device shown can be a software unit, a hardware unit, or a combination of software and hardware built into an existing terminal device. It can also be integrated into the terminal device as an independent component, or it can exist as a standalone terminal device.

[0125] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments 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. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0126] Figure 5 This is a schematic diagram of the robot provided in an embodiment of this application. Figure 5 As shown, the robot 5 in this embodiment includes: at least one processor 50 ( Figure 5 (Only one is shown in the diagram) a processor, a memory 51, and a computer program 52 stored in the memory 51 and executable on the at least one processor 50, wherein the processor 50 executes the computer program 52 to implement the steps in any of the robot control method embodiments described above.

[0127] The processor 50 can be a Central Processing Unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0128] In some embodiments, the memory 51 may be an internal storage unit of the robot 5, such as a hard disk or memory of the robot 5. In other embodiments, the memory 51 may be an external storage device of the robot 5, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the robot 5. Furthermore, the memory 51 may include both internal storage units and external storage devices of the robot 5. The memory 51 is used to store operating systems, applications, boot loaders, data, and other programs, such as the program code of computer programs. The memory 51 can also be used to temporarily store data that has been output or will be output.

[0129] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps in the above-described method embodiments.

[0130] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments.

[0131] 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, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a device / terminal equipment, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0132] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0133] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0134] In the embodiments provided in this application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, 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 or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0135] 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 network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0136] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A robot control method characterized by, include: Obtain the robot's instruction sequence corresponding to the current control moment to obtain the first instruction sequence; The safety confidence level corresponding to the first instruction sequence is calculated based on the preset first feature information; wherein, the first feature information is used to characterize the critical state of the robot's dangerous behavior; and the safety confidence level is used to characterize the safety level of the robot behavior caused by the first instruction sequence. The first instruction sequence is filtered according to the security confidence level to obtain the second instruction sequence; The robot is controlled according to the second instruction sequence.

2. The robot control method as described in claim 1, characterized in that, The step of calculating the security confidence level corresponding to the first instruction sequence based on preset first feature information includes: Obtain the state data corresponding to the current control moment; wherein, the state data includes the state information of the environment in which the robot is located; The second feature information is obtained by fusing and encoding the first instruction sequence and the state data; The security confidence level is calculated based on the first similarity between the first feature information and the second feature information.

3. The robot control method as described in claim 2, characterized in that, The step of calculating the security confidence score based on the first similarity between the first feature information and the second feature information includes: Calculate the first similarity between the first feature information and the second feature information; Obtain the second similarity corresponding to each historical moment within the historical time window prior to the current moment; wherein, the second similarity corresponding to the historical moment is the similarity between the first feature information and the third feature information, and the third feature information is feature information generated based on the robot's instruction sequence and state data corresponding to the historical moment; The first similarity and the second similarity corresponding to each historical moment are weighted and summed according to the preset weights corresponding to each historical moment to obtain the third similarity; wherein, the preset weights are used to characterize the degree of time decay; The security confidence level is determined based on the third similarity.

4. The robot control method as described in claim 3, characterized in that, Determining the security confidence level based on the third similarity includes: The third similarity is nonlinearly mapped according to preset parameters to obtain the security confidence score; The preset parameters include a first parameter and a second parameter. The first parameter is used to characterize the sensitivity to dangerous behavior, and the second parameter is used to characterize the steepness of the critical state from safety to danger.

5. The robot control method according to any one of claims 2 to 4, characterized in that, The step of filtering the first instruction sequence according to the security confidence level to obtain the second instruction sequence includes: The security status level corresponding to the first instruction sequence is determined based on the security confidence level. The first instruction sequence is filtered according to the filtering strategy corresponding to the security status level to obtain the second instruction sequence.

6. The robot control method as described in claim 5, characterized in that, The step of filtering the first instruction sequence according to the filtering strategy corresponding to the security state level to obtain the second instruction sequence includes: If the security status level is the first level, then the first instruction sequence is recorded as the second instruction sequence; If the security status level is the second level, then the first instruction sequence is subjected to overload constraints to generate the second instruction sequence; If the security status level is level three, then the second instruction sequence is set to empty; The security confidence level corresponding to the first level is higher than that corresponding to the second level, and the security confidence level corresponding to the second level is higher than that corresponding to the third level.

7. The robot control method as described in claim 6, characterized in that, The step of overload-constraining the first instruction sequence to generate the second instruction sequence includes: The first instruction sequence is proportionally adjusted based on the security confidence level to obtain the adjustment instruction; The adjustment instructions are optimized using preset rules as constraints to obtain the second instruction sequence; wherein the preset rules are used to describe the safety boundary conditions during the robot's movement.

8. The robot control method as described in claim 1, characterized in that, The method further includes: Acquire scene information for various hazardous scenarios; wherein, the scene information for each hazardous scenario includes a parameter dictionary and a topology map of the hazardous scenario; the parameter dictionary of the hazardous scenario includes text information describing the robot's behavior in the hazardous scenario and corresponding physical control parameters; the topology map of the hazardous scenario includes the spatiotemporal relationship between the robot and other objects in the environment in the hazardous scenario; Obtain preset physical boundary information; wherein, the physical boundary information is used to describe the physical boundary conditions during the robot's movement. The first feature information is generated based on the physical boundary information and the scene information of each of the various dangerous scenarios.

9. A robot comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 8.

Images