84 results about "Motion space" patented technology

The invention provides an autonomous navigation path planning method, device and equipment for a humanoid robot in a complex environment. A navigation task is determined by obtaining the current position and path requirement of the robot; constructing a constraint preposition map based on body size constraint, joint motion constraint and dynamic balance constraint, and calculating a feasible motion space of the robot in advance; acquiring environment sensing data for variation trend analysis, and predicting a future movement track of the dynamic obstacle to generate a dynamic environment prediction sequence; establishing a bidirectional planning mechanism, performing forward and reverse path search at the same time, and generating a plurality of candidate path hypotheses through meeting point detection and path fusion; analyzing a human behavior pattern to identify a potential passage conflict area, establishing a social negotiation strategy, and taking an intermediate meeting point as a social key node to adaptively adjust a candidate path; the optimal safety path is determined through multi-dimensional safety evaluation of the falling risk, the collision risk and the energy consumption risk, and intelligent autonomous navigation of the humanoid robot in a complex environment is achieved.

The invention relates to the technical field of robot joint control, and discloses a robot dog mechanical joint control method. The method comprises the following steps: acquiring a whole-body joint real-time sensing data set which covers the torque reading of a multi-axial force sensor, the angle data of a joint encoder and the body attitude data of an inertial measurement unit; motion state compensation is carried out on the sensing data based on a dynamic window mechanism, and a dynamic compensation matrix containing kinematics compensation parameters is generated; and inputting the multi-modal motion control data set into a gradient descent model under space-time constraint to complete multi-modal motion feature fusion so as to obtain a fusion motion control data set. Joint behavior abnormity is detected through a Lyapunov stability analysis algorithm, and abnormal nodes are identified; and iteratively optimizing control parameters through a fuzzy PID optimization algorithm to generate an optimized control parameter set, finally constructing a three-dimensional virtual motion space model, and establishing a dynamic mapping relationship between a virtual track and an actual mechanism.

According to the intelligent obstacle avoidance reinforcement learning algorithm for the robot in the narrow area, more than ten kinds of common narrow terrains are added in a training map, path points are arranged on the two sides of a target terrain, and a reward function with the robot accurately penetrating through the path points as a core is designed; the problems of high complex terrain collision rate and path planning failure caused by model approximation and single training environment in the prior art are solved. Depth camera data is compressed into a 56 * 80-dimensional feature matrix, ultrasonic sensor data is extracted into a 12-dimensional distance vector, multi-modal data fusion training is performed through a three-layer perceptron network, and a 6-dimensional continuous control quantity including three translational speeds and three angular speeds is directly output. The problems of inaccurate size constraint perception and rough action space caused by dependence on a simplified model in a traditional algorithm are solved, refined control over six-degree-of-freedom motion of the robot in a narrow area is achieved, and the obstacle avoidance safety and the path passing rate in a complex obstacle environment are improved.

The invention discloses a remote humanoid mapping control method based on a virtual reality motion capture mechanical arm. A motion data acquisition module transmits a matrix of a head coordinate system relative to a world coordinate system and a homogeneous transformation matrix to a coordinate system conversion module, establishes a base coordinate system and a tail end coordinate system, and establishes a kinematic chain conversion model; extreme position data of forearm extension of a user is collected, a proportionality coefficient is calculated, and a motion space boundary of the mechanical arm is output; pose normalization processing is carried out to form a discretized mechanical arm tail end pose sequence, and then the spatial relation between the mechanical arm sphere model and the obstacle is analyzed; generating a collision-free track by adopting a parallelization gradient optimization algorithm; the trajectory optimization module realizes trajectory smoothness and time optimization by constructing a multi-objective optimization function; and a mechanical arm motion simulation system is constructed, the tail end track precision is verified, and the controller drives the mechanical arm to achieve high-fidelity reproduction of user actions according to the track. According to the invention, high-precision and smooth-path motion migration of the mechanical arm is realized.

The invention relates to a mechanical arm trajectory optimization method and system based on adversarial learning, and specifically, the method comprises the steps: discretizing a motion space of a mechanical arm to obtain voxels, obtaining a multi-channel image based on whether the voxels contain obstacles or not, and obtaining a to-be-optimized voxel sequence of the mechanical arm through an RRT algorithm; voxel sequences under different conditions are calculated by adopting different neighborhood radiuses and step lengths; calculating a label of a voxel sequence according to the neighborhood radius and the step length, extracting features of the multi-channel image through a feature extraction unit of a discriminator, performing feature fusion on the features of the multi-channel image and the voxel sequence to obtain output of the discriminator, and pre-training the discriminator based on the label and the output of the discriminator; and training the adversarial network by adopting a pre-trained discriminator, and inputting the multi-channel image and the voxel sequence to be optimized into the trained generator to obtain a mechanical arm trajectory optimization result.

The invention relates to the technical field of visual real-time detection and evaluation of fitness actions, and particularly discloses a barbell fitness action real-time detection and evaluation method, which comprises the following steps of: determining a barbell fitness action type, acquiring an image, inputting a network model to predict and obtain a barbell rod area coordinate, mapping to obtain depth information, and extracting point cloud data of a barbell rod area; and fitting a barbell rod center straight line to obtain a real-time motion space trajectory of the barbell, analyzing and calculating to obtain barbell rod motion change information in a motion period, and analyzing and evaluating with a pre-created motion layering condition library to obtain real-time detection and evaluation information. The method combines a deep learning algorithm and a depth camera, can effectively identify and process, has high-precision and high-accuracy detection and evaluation effects, and is suitable for detection and evaluation of various actions of barbell fitness exercises.

The invention belongs to the field of autonomous operation of underwater robots, and particularly relates to a visual mechanical arm grabbing method after an autonomous underwater robot sits on the bottom, which comprises the following steps: after the AUV sits on the bottom, a binocular camera is utilized to shoot seabed images, and target identification and positioning of seabed grabbing are carried out; judging whether a grabbed target is in a seabed motion space of the flexible hand or not by utilizing a transformation matrix of a camera coordinate system and a mechanical arm coordinate system, if not, performing AUV body pose adjustment, and re-executing target recognition and positioning; and if the target is in the seabed motion space of the flexible hand, the AUV performs mechanical arm motion planning and execution based on a table look-up interpolation method according to the spatial position information of the target, so that the flexible hand moves to the target position, and the flexible hand is controlled to be closed to complete grabbing. According to the grabbing motion planning method for the visual mechanical arm, autonomous grabbing actions of the mechanical arm and the flexible hand can be rapidly completed under the low calculation complexity, the grabbing position error is small, and the grabbing success rate is high.

The present invention discloses a method, system, and medium for detecting the spatial posture error of a moving component. The method comprises the following steps: obtaining posture information of a machine tool moving component based on an IMU module and an electronic level: accelerometer signals and angular velocity signals obtained by the IMU module, and angular error signals obtained by the electronic level; fusing the posture information to obtain linear error and angular error; establishing an error prediction model using a BP neural network based on the linear error, angular error, and the motion speed of the machine tool linear axis; obtaining the optimal initial weights and thresholds of the BP neural network using a longhorn beetle whisker search algorithm, substituting these values ​​into the error prediction model for model training to obtain a trained error prediction model; and performing spatial posture error detection on the moving component based on the trained error prediction model to obtain posture error detection results. The present invention can measure the spatial posture error of equipment motion and improve measurement precision while ensuring accuracy.

The invention relates to the technical field of time delay measurement, and provides a robot control end-to-end time delay measurement method and device and electronic equipment. The method comprises the following steps: acquiring a trigger timestamp of a robot motion instruction; based on a candidate displacement composite search strategy and a cost function early stop mechanism, performing global motion estimation on a monitoring video stream corresponding to the robot to obtain a global timestamp of global motion of a monitoring image in the monitoring video stream; the candidate displacement composite search strategy is used for adjusting a search sequence of candidate displacement between adjacent monitoring images based on motion space-time continuity; under the condition that the candidate displacement triggers a cost function early stop mechanism, the cost function early stop mechanism is used for terminating calculation of a cost function of the candidate displacement in advance; and determining an end-to-end time delay corresponding to the robot based on the global timestamp and the trigger timestamp. According to the embodiment of the invention, the end-to-end time delay of a complete link can be measured, and the accuracy and real-time performance of time delay measurement are improved.

The present invention provides a control system for a rehabilitation robot, the method comprising: constructing a motion control diagram according to the real-time collected motion data of the rehabilitation robot and the patient's upper limbs; calculating the displacement data of the end effector according to the motion control diagram and the kinematic model of the rehabilitation robot; establishing the maximum motion space of the end effector according to the kinematic model of the equalization mechanism of the rehabilitation robot, the displacement data of the end effector and the kinematic model; calculating the effective working space of the rehabilitation robot using the end effector and the maximum motion space of the upper limb hemiplegic patient, and controlling the rehabilitation robot using the effective working space. The present invention calculates the displacement data of the end effector using the kinematic model and the motion control diagram, and establishes the mechanical model of the rehabilitation robot using the dynamic model and the displacement data, thereby realizing the control of the rehabilitation robot.

The embodiment of the invention discloses a path planning method, a mobile robot and related equipment, and is applied to the technical field of path navigation. According to the embodiment of the invention, the method comprises the steps: obtaining a plurality of planning paths of the mobile robot, and discretizing a motion space where the plurality of planning paths are located into a plurality of cells; determining first position information of the cells in the motion space and first timestamp information when the planned path arrives at the cells; predicting second position information of cells passed by the dynamic obstacle in the motion space and second timestamp information when the dynamic obstacle arrives at the cells; the dynamic barrier is a position-changing barrier; based on the first position information, the second position information, the first timestamp information and the second timestamp information, determining a collision path colliding with the dynamic obstacle; determining a target planning path from the plurality of planning paths based on the collision path; and the path planning speed can be effectively improved.

This invention relates to a humanoid robot, specifically a parallel rotating mechanical joint based on a ring-shaped guide rail, mimicking the rotation of a human neck. The aim is to provide a parallel rotating mechanical joint based on a ring-shaped guide rail to achieve three degrees of freedom in the movement of a biomimetic human neck joint while ensuring high rigidity and motion accuracy. Furthermore, the joint features a simplified structure, good support performance, and sufficient movement space for the neck joint. The technical solution is a parallel rotating mechanical joint based on a ring-shaped guide rail, characterized by: a head motion platform, a neck ring-shaped guide rail base, and three PSS (Personal Stability Support) link branches connected in parallel between the head motion platform and the neck ring-shaped guide rail base; the center of the head motion platform and the center of the neck ring-shaped guide rail base are connected by a composite rotating shaft.

The invention discloses a radiation treatment bed anti-collision method and system, and the method comprises the steps: drawing a 3D envelope boundary for constraining the motion space of a treatment bed; establishing an incidence relation model among the position of the treatment bed in the 3D envelope boundary, the motor operation current variable quantity and the displacement variable quantity; in the radiotherapy process, the motion space position of the radiotherapy bed and the motor operation current variable quantity and the displacement variable quantity corresponding to the motion space position are monitored in real time, and whether the treatment bed has the collision risk or not is judged based on the incidence relation model. According to the scheme, automatic collision prevention is achieved, when collision occurs, the treatment bed stops moving immediately, and a collision early warning mechanism is provided.

The invention belongs to the technical field of medical robots, and particularly relates to a puncture surgery robot system. The CT bed is provided with the multi-axis movement mechanism, the multi-axis movement mechanism comprises an X-axis linear guide rail mechanism, a Y-axis linear guide rail mechanism, a Z-axis linear guide rail mechanism and a tail end executing mechanism, a cantilever structure in the tail end executing mechanism is connected with the Z-axis linear guide rail mechanism, and a first rotating shaft arranged in the tail end executing mechanism is connected with a second rotating shaft through an output connecting rod. The axis where the first rotating shaft is located and the axis where the second rotating shaft is located are arranged in a crossed mode. The method has the advantages that the positioning error is smaller than or equal to 0.3 mm, and the angle adjusting precision reaches 0.1 degree; non-coplanar puncture paths of anatomical structures such as lungs and livers are adapted; 180-degree horizontal rotation and + / -90-degree pitching adjustment are achieved, the movement space limitation of a parallel structure is broken through, and the device is compatible with obese patients and special body positions such as side lying and prone lying; target point coordinates are fitted in real time through CT images, the motion trail of the robot is dynamically corrected, the manual calibration frequency is reduced, and the operation efficiency is greatly improved.

The application discloses a motion planning method and equipment for a steam generator tube plate quadruped robot, and belongs to the technical field of motion planning of a nuclear industry steam generator quadruped robot. In order to solve the problems that the existing robot has low planning efficiency when inspecting a steam generator tube plate, and the execution success rate of a planning scheme needs to be improved, the application firstly performs traversal search to determine the motion space of a base and a foot end, and determines a feasible foot landing set; then divides a given polyline path into multiple straight line paths, plans a base track, determines the base position, and for each straight line path, the robot selects a foot landing point in the feasible foot landing set according to a turning motion evaluation index, continuously performs a turning motion process until the robot adjusts the posture to face a given orientation and completes the turning motion; the robot selects a foot landing point in the feasible foot landing set according to a straight line motion evaluation index, continuously repeats a straight line motion process until the base of the robot reaches the terminal point of the straight line path.

The application belongs to the field of autonomous operation of underwater robots, and particularly relates to a visual mechanical arm grabbing method after autonomous underwater robots sit on the bottom, which comprises the following steps: after the AUV sits on the bottom, a binocular camera is used to shoot the seabed image, and target recognition and positioning for seabed grabbing are performed; a transformation matrix of the camera coordinate system and the mechanical arm coordinate system is used to determine whether the grabbing target is in the seabed movement space of the flexible hand, if not, the AUV body pose is adjusted, and the target recognition and positioning are re-executed; if the target is in the seabed movement space of the flexible hand, the AUV performs mechanical arm movement planning and execution based on the look-up table interpolation method according to the spatial position information of the target, so that the flexible hand moves to the target position, and the flexible hand is controlled to close to complete grabbing. The visual mechanical arm grabbing movement planning method provided by the application can quickly complete the autonomous grabbing action of the mechanical arm and the flexible hand under low calculation complexity, and has small grabbing position error and high grabbing success rate.

The invention provides a robot mechanical arm motion space simulation method and system and a photovoltaic robot. The method comprises the steps that a mechanical arm motion model is constructed according to the size of each joint connecting rod of a mechanical arm and the allowable rotation range of a joint angle; taking an XZ plane or a YZ plane as a display plane of the simulation system, adding a photovoltaic array model on the display plane, taking an area where the photovoltaic array model is located as a target area, and selecting a plurality of position points in the target area as target position points; the target position points are sequentially input into a mechanical arm motion model for motion simulation, and whether the tail end of the mechanical arm can reach all the target position points or not is judged; a point cloud mark is generated at a target position point where the tail end of the mechanical arm can reach, and a first point cloud set is formed; and rotating the first point cloud set around the Z axis of the mechanical arm base by one circle to form a motion space of the mechanical arm. According to the scheme, the motion space of the mechanical arm of the photovoltaic cleaning robot can be simulated, so that design and engineering verification of the mechanical arm are facilitated.

The present invention provides a method and system for calculating the degrees of freedom of a multi-loop coupling mechanism based on geometric algebra, including: determining the basic branch of a Twin-Bennett mechanism to obtain the initial motion space of the basic branch; finding all coupled branches that form a closed loop with the basic branch and calculating the branch motion space of the coupled branch; judging whether the motion space of the coupled branch in the closed loop belongs to the motion space of the basic branch in the closed loop among all basic branches containing the closed loop; if so, equating the motion space of the basic branch in the closed loop to the motion space of the closed loop; if not, calculating the intersection of the motion spaces of the basic branch and the coupled branch in the closed loop to be equivalent to the motion space of the new basic branch; finding the intersection of the motion spaces of all branches of the equivalent mechanism to obtain the motion space of the output platform, thereby deriving the degrees of freedom of the Twin-Bennett mechanism. The present invention has a simple analysis and calculation process and clear physical meaning.

The present invention provides a master-slave isomorphic robot teleoperation safety control method and system. The method collects an environmental image of the slave robot's workspace and determines whether there are obstacles. If there are obstacles, the method calculates the expected acceleration of the slave robot's end caused by virtual gravity and repulsion and maps it to the joint space to obtain the expected angular acceleration of each joint, and then maps the expected angular acceleration to the joint space. Simultaneously, the expected acceleration of the slave robot's mobile chassis caused by virtual gravity and repulsion is calculated and applied to the motion space. If there are no obstacles, the expected angular acceleration of each joint of the slave robot is calculated and applied to the joint space of the slave robot. Simultaneously, the expected acceleration of the slave robot's mobile chassis caused by virtual gravity is calculated and applied to the motion space. By combining robot teleoperation with obstacle avoidance, the method allows the slave robot to safely avoid obstacles while stably following the motion trajectory of the master robot, achieving the goal of synchronized motion between the master and slave robots.

The invention relates to a method for controlling at least one closing movement of a door device (1) and a door device (1) with at least one movable door leaf (2, 3) for a public passenger transport vehicle, in particular for a rail vehicle, comprising the following method steps: • Closing the door device (2, 3) by moving at least one door leaf (2, 3) at a predetermined first speed; • Monitoring at least a partial area (8, 9) of a movement space (4) of the at least one door leaf (2, 3) and, upon detection of an object of at least one predetermined type, shape and / or size in the monitored movement space (4, 8, 9): closing the door device (1) by moving the at least one door leaf (2, 3) with a predetermined second speed, wherein the second speed is lower than the first speed.

The present application relates to the technical field of mechanical motion control, and discloses a trajectory control method and system for a mechanical arm, an electronic device and a storage medium, wherein the trajectory constraint of the mechanical arm is determined according to the motion space boundary limitation of the mechanical arm; the motion trajectory of the mechanical arm is determined through the spherical parameter equation of the mechanical arm; under the trajectory constraint of the mechanical arm, the motion trajectory of the mechanical arm is executed, and the error between the end posture of the mechanical arm and the preset expected posture is minimized, so that the motion trajectory of the mechanical arm can be accurately controlled to a large extent.

The application discloses a kind of five degrees of freedom robot motion control methods, based on discretization motion space, specifically includes the following steps: the motion space of five degrees of freedom robot is divided into multiple regions, and the point in each region is discretized and handled, generates discrete point dataset;Discrete point dataset and its corresponding motion parameter are stored in database, form pose query table;According to the coordinates of target point, find the nearest discrete point in discrete point dataset, and obtain its corresponding motion parameter;The motion parameter found is sent to robot control system, drives robot to complete pose adjustment.The application provides a kind of five degrees of freedom robot motion control method, realizes quick pose query and control, significantly improves the efficiency, accuracy and safety of motion control;The method is suitable for robot control in magnetic resonance interventional surgical treatment, can significantly improve the calculation speed, accuracy and safety, reduce the risk of operation.

The embodiment of the invention provides a humanoid robot action parameter determination method and related equipment, and belongs to the technical field of robot control. The method comprises the following steps: extracting multi-modal input features; in the aerial view space, to-be-observed features generated by fusing the multi-modal input features are obtained, and static parameters of static objects and dynamic parameters of dynamic objects around the humanoid robot are determined; generating a first motion parameter of the humanoid robot based on the to-be-observed features; and determining a motion correction value by using the motion residual value, and modifying the first motion parameter by using the motion correction value to obtain a second motion parameter. According to the technical scheme, the feature information around the humanoid robot is obtained in real time, the dynamic parameters of the dynamic object and the static parameters of the static object are determined in a decoupling mode, the movable space of the mechanical arm is determined, dependence on positioning data is reduced, the obstacle avoidance success rate of the humanoid robot is increased, and the obstacle avoidance efficiency is improved. And the industrial operation efficiency and the household interaction safety of the humanoid robot are effectively improved.

The present application provides a video processing method and related electronic equipment, the method comprising: capturing a first image of a user through a camera; obtaining the first skeletal node coordinates and the first motion space radius of the user in the first image; calculating the target coordinates of the AR element in the i-th frame image according to the first skeletal node coordinates, the first motion space radius, the second skeletal node coordinates, the second motion space radius and the coordinates of the AR element in the i-th frame image of the first video; the i-th frame image is the next frame image to be played of the first video; the i-th frame image is displayed on a display screen; wherein the coordinates of the AR element in the displayed i-th frame image are the target coordinates. Through the above method, the electronic device will adjust the position of the AR element in the first video according to the body proportions of the user, so as to be more adapted to the body proportions of the user, thereby improving the learning effect of the user in learning the actions in the first video based on the AR element.

The invention discloses a lightweight wheelchair auxiliary tendon mechanical arm system based on upper limb posture control, and relates to the technical field of rehabilitation auxiliary robots, the lightweight wheelchair auxiliary tendon mechanical arm system comprises a linear drive seven-degree-of-freedom modular tendon mechanical arm, a standardized mobile wheelchair, an Xsens MTw Awindow inertial measurement unit and a hierarchical control architecture; according to the mechanical arm, through seven-degree-of-freedom modular tendon transmission, light weight and large motion space of the mechanical arm are achieved, wheelchair adjustment is reduced through functional area division, wheelchair rollover can be prevented after integration of L-shaped steel parts, and the conflict between the motion space and the weight and the size is solved; an Xsens inertial sensor is adopted to collect the posture of the upper limb, Kalman filtering and ROS hierarchical control are matched, an operator can accurately complete the tasks of grabbing, cleaning and the like by means of natural upper limb actions without training, and the learning cost is reduced; the mechanical arm is modularly designed to have a replaceable end effector, a control framework is based on ROS, subsequent function expansion or integration to other platforms is facilitated, and compatibility is high.

The invention relates to a mechanical arm collision detection method and device, equipment and a program product, and the method comprises the steps: layering a motion space where a working mechanical arm and an obstacle mechanical arm possibly collide, and obtaining a layering result; based on the layering result, layering spaces to which a plurality of first movement modules of the working mechanical arm and a plurality of second movement modules of the obstacle mechanical arm belong are determined; respectively creating a first collision module corresponding to each first motion module, and respectively creating a second collision module corresponding to each second motion module; and collision detection is conducted on the first collision module and the second collision module which are located in the same layered space, and the collision detection result of the working mechanical arm and the obstacle mechanical arm is obtained. By adopting the method, the collision detection efficiency of the mechanical arm can be improved.

The present invention belongs to the field of infrared imaging technology, and particularly relates to a method for generating a spatially variant blurred image of rotational motion, which is used to solve the problem of simulating and generating blurred images in the case of rotational motion in an imaging system, and provides support for the analysis and demonstration of various infrared imaging systems. The method proposes a method for generating a motion blurred image with spatially variant characteristics when there is rotational motion in an infrared imaging system. A blur kernel is generated for each pixel, and the template direction of the conventional motion blur model is optimized. The method specifically includes: establishing an image coordinate system, generating a rotational motion blur template, obtaining all blurred pixels based on the rotational motion blur template, and thus generating a final rotational motion blurred image. The present invention gives a specific description of the method for generating the blurred image of rotational motion, which can provide support for the analysis and demonstration of various infrared imaging systems.

The invention discloses an XR somatosensory large-space grid platform combined control device and method. The device is composed of a dynamic platform module, a driving control module, a central control module, a multi-sensory simulation system and a VR linkage module. The core characteristic is that a traditional large-area fixed partition plate is replaced by a grid system composed of at least two independent flat platforms. Each platform is provided with an independent servo drive and can shake, lift and vibrate. The multi-sensory system integrates water spraying, smell, temperature and wind power simulation. After a game instruction is received, a four-step control process of mode selection, parameter configuration, synchronous execution and feedback regulation is implemented. The system supports a'full sense mode '(enabling all dynamic and sensory simulations) and a'common mode' (providing a basic load bearing function as a static platform). The platform combination scheme supports multiple forms such as a single platform, double platforms and four platforms, and flexible switching is achieved through software parameters.

The invention discloses an assessment method for the ability of medical personnel to repeatedly complete specified actions, and relates to the field of surgical skill assessment, and the method comprises the steps: collecting a standard path corresponding to a specified action and a plurality of training paths; performing standardization processing on the collected data, calculating distances between all path points in each training path and corresponding path points in a standard path, and sorting the distances from large to small to obtain a second distance set; based on each distance value dj in the second distance set, the volume of a cylinder with the standard path as the center axis is calculated to serve as the motion space volume Vj; sequentially traversing the dj in the second distance set, obtaining a corresponding reliable path number Tj based on each dj, and calculating a corresponding reliability Rj in combination with the total number of training paths; and based on the Rj and the Vj corresponding to each dj, calculating a reliability degradation speed RDVj corresponding to each dj, and based on the RDVj, evaluating the ability of the medical personnel to repeatedly complete the specified action. According to the invention, the ability of medical personnel to repeatedly complete specified actions is quantified.

This application provides a training method, control method, and device for a game strategy network, relating to the field of robot control technology. The method includes: using a robot adversarial action dataset to acquire sample observation data of each preset adversarial action and the task observation state; training an initial motion tracker based on the sample observation data to obtain a target motion tracker; using the target motion tracker to train an initial latent motion control network to obtain a target latent motion control network; training an initial residual policy network against a stationary target within a preset multidimensional latent motion space corresponding to the target latent motion control network to obtain a target residual policy network; and training an initial self-game network in the multidimensional latent motion space based on the warm-up adversarial policy corresponding to the target residual policy network to obtain the target game strategy network. This application improves the accuracy and safety of robot control.