Mobile control method, electronic device, readable storage medium and program product

Abstract

The application relates to the technical field of robot control, and discloses a mobile control method, an electronic device, a readable storage medium and a program product. The method comprises the following steps: detecting a target motion instruction used for indicating the movement of the electronic device, the target motion instruction corresponding to a target position; detecting the terrain features of the current position of the electronic device; generating a target planning track of the electronic device based on the target motion instruction, the terrain features and real-time state parameters of the electronic device; determining the target joint torque of the electronic device based on the target planning track and the real-time state parameters; and controlling the movement of the electronic device by using a controller based on the target joint torque, so that the electronic device can be moved to the target position. In this way, the electronic device can adapt to various terrains, and the movement process is stable and reliable.

Description

Technical Field

[0001] This application relates to the field of robot control technology, and in particular to a motion control method, electronic device, readable storage medium, and program product. Background Technology

[0002] Current robot autonomous navigation control strategies mainly fall into two categories. The first involves calculating a trajectory sequence using pure kinematic programming (a geometric calculation method), and then adjusting joint torques in real time based on the position of each trajectory point to control the robot's movement along the trajectory. The second involves using a deep learning model to directly output joint torque adjustment commands based on perceived environmental information, thereby controlling the robot's movement in real time.

[0003] However, the above control strategies usually assume that the robot is on a flat surface. When the robot is on an uneven surface, the above strategies cannot dynamically determine the appropriate joint torque according to the terrain undulations, which can easily cause the robot to become unstable and fall. Summary of the Invention

[0004] To address the aforementioned problems, this application provides a mobile control method, an electronic device, a readable storage medium, and a program product.

[0005] In a first aspect, embodiments of this application provide a motion control method applied to a movable electronic device. The method includes: detecting a target motion command of the electronic device, the target motion command corresponding to a target position; detecting terrain features at the current location of the electronic device; generating a target planned trajectory of the electronic device based on the target motion command, terrain features, and real-time state parameters of the electronic device; determining a target joint torque of the electronic device based on the target planned trajectory and real-time state parameters; and controlling the electronic device to move to the target position via a controller based on the target joint torque.

[0006] It is understood that in some embodiments of this application, a movable electronic device may refer to a movable robot, etc., and this application does not specifically limit it.

[0007] It is understood that the target motion command can refer to the speed command mentioned in the embodiments of this application for instructing the movement of an electronic device. The target motion command can include the lateral speed, longitudinal speed, and yaw rate of the electronic device. For example, in the embodiments of this application, m_cmd = (Vx, Vy, Vyaw), where m_cmd represents the target motion speed of the electronic device, Vx represents the longitudinal speed of the electronic device, Vy represents the lateral speed of the electronic device, and Vyaw represents the yaw rate of the electronic device. Terrain features can refer to Et mentioned in the embodiments of this application, and terrain features can represent the geometric features of the environment in which the electronic device is located.

[0008] Based on the above method, regardless of the terrain the electronic device is on, it can dynamically generate a target trajectory adapted to the current terrain and calculate the corresponding target joint torque by combining terrain features, target motion commands, and the real-time status parameters of the electronic device, thus achieving adaptive and stable control. This method deeply integrates terrain feature information to complete dynamic trajectory correction, ensuring stable and high-frequency action execution by the equipment. It breaks through the limitation of traditional control strategies that are only applicable to flat roads. Through adaptive adjustment of trajectory and joint torque, it prevents imbalance problems caused by terrain incompatibility, effectively improving the environmental adaptability and operational safety of navigation control.

[0009] In one possible implementation of the first aspect above, the target motion command includes the longitudinal velocity, lateral velocity, and yaw rate of the electronic device; the target motion command is determined based on navigation target information, wherein the navigation target information includes a target position image and / or target position coordinates, the target position image being used to characterize the semantic information of the target position, and the target position coordinates being used to characterize the geometric information of the target position.

[0010] It is understood that the scene target can refer to the target location image mentioned in the embodiments of this application, which is used to represent the semantic information of the target location. The global pose can refer to the target location coordinates (i.e., the pose in the world coordinate system) mentioned in the embodiments of this application, which is used to represent the geometric and orientation information of the target location.

[0011] In one possible implementation of the first aspect mentioned above, the method for determining the terrain features includes: acquiring the geometric features of the environment in which the electronic device is located, inputting the geometric features into a first neural network for feature compression, and obtaining terrain features in the form of latent variables.

[0012] It is understood that the first neural network may refer to the shallow three-dimensional convolutional neural network mentioned in the embodiments of this application, such as shallow 3D CNN, etc., and this application does not make specific limitations on it.

[0013] In one possible implementation of the first aspect above, the electronic device includes multiple moving joints; the real-time state parameters of the electronic device include at least one of the following: the position, orientation, linear velocity, and angular velocity of the center of mass of the electronic device, and the joint position, velocity, rotation angle, and angular velocity of each moving joint.

[0014] In some embodiments of this application, the electronic device may include multiple movable joints. For example, if the electronic device is a humanoid robot, the movable joints may include hip joints, knee joints, ankle joints, etc., and this application does not specifically limit them.

[0015] It can be understood that the center of mass is used to characterize the overall center of gravity position of the electronic device, the orientation is used to characterize the direction of movement of the electronic device, the linear velocity is used to characterize the speed at which the center of mass moves, the angular velocity is used to characterize the rotation rate of the electronic device, and the rotation angle is used to characterize the degree of bending of the moving joint.

[0016] In one possible implementation of the first aspect described above, the target planning trajectory of the electronic device is generated based on the target motion command, terrain features, and real-time state parameters of the electronic device. This includes: concatenating the target motion command, terrain features, and real-time state parameters to obtain a condition vector; processing the time step information and condition vector based on a normalization mechanism to generate scaling parameters and offset parameters for feature extraction, and inputting the scaling parameters and offset parameters into a second neural network; wherein the time step information represents the current time point, and the scaling parameters and offset parameters are used to adjust the condition vector features extracted by the second neural network to obtain adjusted condition vector features; performing vector field prediction on the adjusted condition vector features based on a flow matching model to obtain the state residual trajectory of the electronic device, wherein the state residual trajectory represents the change in the state parameters of each trajectory point relative to the real-time state parameters; and determining the target planning trajectory of the electronic device based on the state residual trajectory and the real-time state parameters.

[0017] In some embodiments of this application, the condition vector may refer to ct=[pt,m_cmd,Et] mentioned in the embodiments of this application, which integrates three types of information: target motion command of electronic device, terrain features and real-time status parameters.

[0018] In some embodiments of this application, the electronic device may employ an adaptive normalization mechanism to fuse and normalize the time step information and conditional vector features to generate corresponding scaling coefficients γ and offset coefficients β. γ and β are then input into a second neural network. The second neural network may adjust the distribution of the extracted conditional vector features based on the scaling coefficients γ and offset coefficients β to obtain adjusted conditional vector features, thereby improving the stability and adaptability of feature extraction under different time steps and conditions.

[0019] It can be understood that the second neural network is the backbone network of the generative model. For example, the second neural network can be a Transformer, used to encode features and model temporal relationships of conditional vectors to obtain adjusted conditional vector features.

[0020] In some embodiments of this application, based on a flow matching model and using adjusted conditional vector features as input, the evolution of the electronic device's state parameters in the continuous time domain is learned, and a smooth vector field pointing from the current real-time state parameters to the target state parameters is predicted. Based on this vector field, a continuous, abrupt state residual trajectory can be generated, ensuring that the robot's state changes conform to dynamic constraints and avoiding problems such as trajectory jumps, jitter, or discontinuities.

[0021] In one possible implementation of the first aspect above, the target motion command, terrain features, and real-time state parameters are concatenated to obtain a condition vector, including: if the terrain features indicate that the location of the electronic device is a non-flat road surface, then the target motion command is adjusted based on the terrain features, and the adjusted target motion command, terrain features, and real-time state parameters are concatenated to obtain a condition vector.

[0022] In some embodiments of this application, during the process of concatenating the target motion command, terrain features, and real-time state parameters to obtain a condition vector, if the terrain features indicate that the location of the electronic device is a non-flat road surface, the target motion command can be adjusted based on the terrain features, and the adjusted target motion command, terrain features, and real-time state parameters can be concatenated to obtain a condition vector.

[0023] In one possible implementation of the first aspect above, determining the target planning trajectory of the electronic device based on the state residual trajectory and real-time state parameters includes: obtaining the initial motion trajectory of the electronic device within a preset time period based on the state residual trajectory and real-time state parameters, performing third-order spline interpolation on the initial motion trajectory to obtain the target planning trajectory, wherein the third-order spline interpolation is used to make the density of the target planning trajectory greater than that of the initial motion trajectory.

[0024] In some embodiments of this application, third-order spline interpolation can significantly improve the sampling point density of the target planned trajectory, making the trajectory continuous and dense in the time dimension, thereby ensuring the continuous and smooth transition of key motion parameters such as attitude, position, and angle, and making the motion of electronic devices more coherent and stable.

[0025] In one possible implementation of the first aspect described above, determining the target joint torque of the electronic device based on the target planned trajectory and real-time state parameters includes: compressing the target planned trajectory using a motion segmenter based on finite scalar quantization to obtain discrete target trajectory semantic tags, wherein the target trajectory semantic tags are used to characterize the motion semantic information of each trajectory point in the target planned trajectory; decoding the target trajectory semantic tags in combination with the historical state parameters of the electronic device to obtain a target joint position matrix, wherein the target joint position matrix is ​​used to characterize the target state parameters of the joints of the electronic device at each trajectory point; and determining the target joint torque matrix of the electronic device based on the target joint position matrix and the real-time state parameters.

[0026] It is understood that the motion segmenter based on finite scalar quantization can refer to the motion segmenter based on improved finite scalar quantization mentioned in the embodiments of this application.

[0027] In some embodiments of this application, the electronic device can extract the state parameter features of each time-series node in the target planned trajectory based on a motion segmenter, and delineate discrete semantic intervals through preset quantization rules, mapping continuous motion state data into standardized semantic units within a finite set, i.e., target trajectory semantic tags. For example, the target trajectory semantic tag corresponding to each trajectory point can be represented as discrete motion mode labels such as smooth straight-line movement, slight uphill climb, steering adjustment, and joint easing.

[0028] In some embodiments of this application, the electronic device can use an actor-critic architecture (actor-critic architecture) to decode the abstract discrete trajectory semantic tags into joint target positions corresponding to each trajectory point. This is achieved by taking discrete target trajectory semantic tags and real-time state parameters as input, relying on the built-in semantic decoding mapping relationship, and combining the constraints of historical state parameters accumulated over time. These abstract discrete trajectory semantic tags are then integrated to form a target joint position matrix. For example, the target joint position matrix is ​​a two-dimensional structured matrix, with rows and columns corresponding to joint numbers and time-series trajectory points, respectively. The joint dimension matches the number of moving joints in the electronic device (e.g., when the electronic device contains 29 moving joints, the joint dimension of the target joint position matrix is ​​29). Each element in the target joint position matrix can specifically represent the target joint position of the corresponding joint at a single trajectory point, as well as the target joint angle, target joint velocity, or target joint angular velocity, etc.

[0029] In some embodiments of this application, the electronic device can calculate the deviation between the real-time joint position of each moving joint and the target joint position based on a (high-frequency) proportional-differential controller combined with the target joint position matrix and real-time state parameters. Based on the deviation results, the joint torque corresponding to each joint is obtained, and these are combined to obtain the target joint torque matrix. The electronic device can then drive the movement of each moving joint of the electronic device using the target joint torque matrix based on the controller, thereby achieving smooth movement of the electronic device.

[0030] Secondly, embodiments of this application provide a mobile control system, which includes a macroscopic command module, a terrain and state condition generation module, and a high-frequency universal full-body tracking module. The macroscopic command module is used to detect target motion commands of an electronic device, and the target motion commands correspond to target positions. The terrain and state condition generation module is used to detect terrain features at the current position of the electronic device, and the terrain and state condition generation module is used to generate a target planned trajectory of the electronic device based on the target motion commands, terrain features, and real-time state parameters of the electronic device. The high-frequency universal full-body tracking module is used to determine the target joint torque of the electronic device based on the target planned trajectory and real-time state parameters, and the high-frequency universal full-body tracking module is used to control the electronic device to move to the target position based on the target joint torque.

[0031] Thus, in this application, the functional modules work in a layered and coordinated manner, deeply integrating terrain features to complete dynamic trajectory correction and ensure stable high-frequency execution of actions by the equipment. Regardless of the terrain in which the electronic device is located, it can dynamically generate a target planning trajectory adapted to the terrain and calculate the target joint torque by combining terrain features, target motion commands, and real-time status parameters, thereby achieving full-domain adaptive stability control.

[0032] In one possible implementation of the second aspect described above, the terrain and state condition generation module includes an environmental encoder, and the high-frequency universal full-body tracking module includes a joint proportional differential controller. The environmental encoder is used to detect the terrain features of the current location of the electronic device; the joint proportional differential controller is used to determine the target joint torque of the electronic device and control the movement of the electronic device based on the target joint torque to move to the target position. Specifically, the macroscopic instruction module operates based on a first frequency, the terrain and state condition generation module operates based on a second frequency, the high-frequency universal full-body tracking module operates based on a third frequency, and the joint proportional differential controller operates based on a fourth frequency; the fourth frequency is greater than the third frequency, the third frequency is greater than the second frequency, and the second frequency is greater than the first frequency.

[0033] In some embodiments of this application, the first frequency can be 1Hz to 5Hz, the second frequency can be 25Hz, the third frequency can be 50Hz, and the fourth frequency can be 200Hz. It is understood that the above frequency values ​​are merely illustrative, and this application does not specifically limit the values ​​of the first, second, third, and fourth frequencies.

[0034] Thus, the macroscopic instruction module can operate at low frequencies of 1Hz to 5Hz to meet the low real-time requirements of global navigation intent generation. The terrain and state condition generation module can operate at a mid-frequency of 25Hz to balance the complexity of terrain feature processing with the timeliness of trajectory updates. The high-frequency general-purpose whole-body tracking module can operate at a mid-to-high frequency of 50Hz to ensure the real-time performance of trajectory decoding and torque planning. The joint proportional differential controller can operate at a high frequency of 200Hz to achieve rapid response and closed-loop control of joint movements, with each level of frequency precisely matching the performance requirements of its corresponding function.

[0035] Thirdly, embodiments of this application provide an electronic device, including: at least one memory and at least one processor, the memory being coupled to the processor, the memory being used to store computer program code / instructions; when the computer program code / instructions are executed by the processor, the electronic device causes the electronic device to implement the motion control method of the first aspect and any of the possible implementations of the first aspect described above.

[0036] Fourthly, embodiments of this application provide a readable storage medium storing instructions that, when executed on an electronic device, cause the electronic device to implement the motion control method described in the first aspect and any of its possible implementations.

[0037] Fifthly, embodiments of this application provide a computer program product, including: computer instructions, which, when executed on an electronic device, cause the electronic device to implement the motion control method described in the first aspect and any of the possible implementations of the first aspect.

[0038] The beneficial effects of the third to fifth aspects mentioned above can be referred to the relevant descriptions in the first aspect and any possible implementation of the first aspect, which will not be repeated here. Attached Figure Description

[0039] Figure 1 According to some embodiments of this application, a flowchart of a motion control method is shown;

[0040] Figure 2 According to some embodiments of this application, a schematic diagram of the result of a mobile control system is shown;

[0041] Figure 3 According to some embodiments of this application, a process schematic diagram of a motion control method is shown;

[0042] Figure 4 According to some embodiments of this application, a schematic diagram of the structure of an electronic device is shown. Detailed Implementation

[0043] The illustrative embodiments of this application include, but are not limited to, a mobile control method, an electronic device, a readable storage medium, and a program product.

[0044] It should be noted that the motion control method provided in this application can be applied to various types of mobile electronic devices, such as mobile robots, humanoid robots, legged robots, wheeled robots, multi-joint electromechanical platforms, and other devices with autonomous motion control capabilities. This application does not limit the specific form of the electronic device.

[0045] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0046] As mentioned earlier, current robot autonomous navigation control strategies cannot dynamically calculate appropriate joint torques based on terrain undulations when the robot is on uneven surfaces, which can easily cause the robot to become unstable and fall.

[0047] To address the aforementioned issues, this application provides a motion control method applied to a mobile electronic device, such as a mobile robot. The method specifically includes: detecting a target motion command for the electronic device, the target motion command corresponding to a target position, to instruct the electronic device to move from its current position to the target position; detecting terrain features at the current location of the electronic device; generating a target planned trajectory for the electronic device based on the target motion command, terrain features, and real-time state parameters of the electronic device; determining a target joint torque for the electronic device based on the target planned trajectory and real-time state parameters; and controlling the electronic device to move to the target position using a controller based on the target joint torque.

[0048] Based on the above method, regardless of the terrain the electronic device is on, it can dynamically generate a target planning trajectory adapted to the current terrain and calculate the corresponding target joint torque by combining terrain features, target motion commands, and the real-time status parameters of the electronic device, thus achieving adaptive stability control. For example, when the current terrain has undulations, potholes, or obstacles, the target planning trajectory can be corrected in a timely manner based on the real-time perceived terrain features and target motion state, so as to calculate the action commands for stepping, obstacle avoidance, or restoring balance, avoiding the device from losing balance and falling due to unsuitable terrain, and improving the environmental adaptability and safety of navigation control.

[0049] This method deeply integrates terrain feature information to achieve dynamic trajectory correction, ensuring stable and high-frequency operation of the equipment. It breaks through the limitation of traditional control strategies that are only applicable to flat roads. Through adaptive adjustment of trajectory and joint torque, it prevents imbalance problems caused by terrain incompatibility, effectively improving the environmental adaptability and operational safety of navigation control.

[0050] In the following embodiments, a mobile electronic device, specifically a robot, is used as an example to describe the motion control method in this application. For instance, according to some embodiments of this application, Figure 1 A flowchart illustrating a motion control method is shown. It should be understood that... Figure 1 The entities executing each process shown can be mobile robots. Specific processes may include the following:

[0051] S101: A target motion command for instructing the electronic device to move is detected, and the target motion command corresponds to the target position.

[0052] In some embodiments of this application, after the robot is started, a target motion command can be determined from the current position to the target position based on the navigation target information. The target motion command is a speed command used to instruct the robot to move. The target motion command may include the robot's lateral speed, longitudinal speed and yaw rate.

[0053] For example, m_cmd = (Vx, Vy, Vyaw), where m_cmd can represent the target motion velocity of the robot, Vx can represent the longitudinal velocity of the robot, Vy can represent the lateral velocity of the robot, and Vyaw can represent the yaw angular velocity of the robot.

[0054] In some embodiments of this application, the robot can determine navigation target information based on a topology map, and then determine target motion commands based on the navigation target information. The navigation target information may include a target location image and / or target location coordinates. Specifically, the robot can construct an environmental visual topology map G(V, E) using point clouds or depth maps collected by visual sensors. Node V can be key locations such as doors, corridor intersections, and room entrances, and edges E can be traversable connections between nodes. The robot can map the target location to a target node vg in the topology map using semantic understanding methods such as keyword matching and visual language models.

[0055] The target location image can be the scene visual data corresponding to the target node vg. This scene visual data can be generated by the robot facing the area where the target node is located, through the fusion of images acquired by the visual camera and LiDAR with point cloud data, which can characterize the semantic features and environmental structure of the target location.

[0056] The target position coordinates can be the pose of the target node vg in the world coordinate system, which can be obtained in two ways: First, when the topology graph is pre-built, the world coordinate system pose of each node is bound to the node and stored in the robot. Second, it is obtained through multi-source data fusion calculation using laser point cloud matching, visual simultaneous localization and mapping (Visual SLAM), and inertial measurement unit. The pose of the target node vg in the world coordinate system can be used to accurately determine the spatial orientation and orientation of the target position in the world coordinate system, and can be represented as (xg, yg, yawg).

[0057] After determining the target location image and target location coordinates, the robot can obtain its global pose (x, y, yaw) in the world coordinate system in real time through the positioning module. Combined with the world coordinates (xg, yg) of the target node vg output by the path planning algorithm, the relative position deviation (Δx, Δy) between the robot and the target node vg is calculated. This relative position deviation is the difference in the world coordinate system. The robot performs coordinate rotation transformation in combination with its own heading angle to convert the deviation data to the robot's body coordinate system. Then, the longitudinal velocity, lateral velocity, and yaw angular velocity are calculated by the proportional controller or kinematic model and combined into a macroscopic motion command m_cmd, which serves as the target motion command for the robot's movement.

[0058] For example, the robot's current pose is (1.00, 1.00, 0.785 rad), the target position coordinates are (3.50, 2.25), the relative deviation is calculated (Δx=2.50, Δy=1.25), and the speed command is output after coordinate transformation (Vx=0.8m / s, Vy=0, Vyaw=0.15 rad / s).

[0059] In some embodiments of this application, the aforementioned target motion command may refer to the desired motion command under ideal conditions. Ideal conditions may refer to a standard traffic environment where the road surface is flat, there are no potholes or steep slopes, there is no external attitude interference, and the aircraft body remains stable. In this scenario, there is no need to impose additional restrictions on the travel speed and steering range, and the output command is the ideal expected value to ensure traffic efficiency.

[0060] S102: Terrain features of the current location of the electronic device have been detected.

[0061] In some embodiments of this application, the robot can acquire the geometric features of its environment (i.e., high-dimensional geometric data of the local terrain), input these geometric features into a first neural network for feature compression, and obtain terrain features in the form of latent variables. The first neural network can refer to a convolutional neural network (CNN), such as a shallow 3D convolutional neural network, i.e., a shallow 3D CNN. Specifically, the robot can use a shallow 3D CNN to perform feature extraction and dimensionality reduction on the input high-dimensional geometric features, compressing them into terrain features Et in the form of low-dimensional latent variables.

[0062] S103: Generate the target planning trajectory of the electronic device based on the target motion command, terrain features, and real-time status parameters of the electronic device.

[0063] In some embodiments of this application, the robot can acquire its state parameters in real time, i.e., the robot's real-time state parameters, and then generate the robot's target planning trajectory based on the target motion command, terrain features and the robot's real-time state parameters.

[0064] In some embodiments of this application, the robot may include multiple movable joints. For example, if the robot is a humanoid robot, the movable joints may include the hip joint, knee joint, ankle joint, etc., and this application does not specifically limit them.

[0065] In some embodiments of this application, the real-time state parameters of the robot may include at least one of the following: the position, orientation, linear velocity, and angular velocity of the robot's center of mass, and the joint position, velocity, rotation angle, and angular velocity of each moving joint.

[0066] It should be understood that the center of mass is used to characterize the overall center of gravity position of the robot, the orientation is used to characterize the robot's forward direction, the linear velocity is used to characterize the speed of the center of mass movement, the angular velocity is used to characterize the rotation rate of the robot body, and the rotation angle is used to characterize the degree of joint bending. For example, the robot's real-time state parameter pt can be expressed as: {center of mass position (x=1.00, y=1.00), orientation yaw=0.785rad, center of mass linear velocity (vx=0.8m / s, vy=0m / s), center of mass angular velocity ω=0.1rad / s; left hip joint position θ1=0.52rad, joint velocity ω1=0.2rad / s; left knee joint position θ2=-1.05rad, joint velocity ω2=-0.1rad / s; position and velocity parameters of the right hip joint, right knee joint, and other joints}. This application does not impose specific limitations on the numerical values.

[0067] In some embodiments of this application, a target planning trajectory for the robot is generated based on the target motion command, terrain features, and the robot's real-time state parameters. Specifically, this may include the following process.

[0068] First, the robot can concatenate the target motion command m_cmd, terrain features Et, and real-time state parameters pt to obtain a conditional vector. For example, the robot can concatenate m_cmd, Et, and pt along the channel dimension to form a one-dimensional continuous feature vector, i.e., the conditional vector ct. For instance, the conditional vector ct can be represented as a 512-dimensional feature vector that integrates the robot's target motion command, terrain features, and real-time state parameters.

[0069] In the process of concatenating the target motion command, terrain features, and real-time state parameters to obtain the condition vector, if the terrain features indicate that the robot's location is on a non-flat road surface, the target motion command can be adjusted based on the terrain features, and the adjusted target motion command, terrain features, and real-time state parameters can be concatenated to obtain the condition vector.

[0070] For example, when a steep slope is detected by the terrain features, the longitudinal velocity can be automatically reduced from 0.8 m / s to 0.3 m / s to ensure movement stability. For instance, the adjusted target motion command can be (Vx=0.3 m / s, Vy=0, Vyaw=0.05 rad / s), and then the adjusted target motion command, terrain features, and real-time state parameters are concatenated to obtain a conditional vector.

[0071] Secondly, the robot can process the time step information and conditional vectors based on a normalization mechanism to generate scaling and offset parameters for feature extraction, and then input these parameters into a second neural network. For example, the robot can employ adaptive instance normalization (AdaIN) to fuse and normalize the time step information and conditional vector features, generating corresponding scaling coefficients γ and offset coefficients β. These γ and β are then input into the second neural network, which can adjust the distribution of the extracted conditional vector features based on the scaling coefficients γ and offset coefficients β to obtain adjusted conditional vector features, thereby improving the stability and adaptability of feature extraction under different time steps and conditions.

[0072] It can be understood that time step information can be represented as a floating-point number between 0 and 1, used to distinguish the order in which the trajectory is generated. For example, t=0.20 indicates that the current time step is the 20th percentile of trajectory generation. The second neural network is the backbone network of the generative model. For example, the second neural network can be a Transformer, used to encode features and model temporal relationships in the conditional vectors to obtain adjusted conditional vector features.

[0073] Then, the robot can perform vector field prediction on the adjusted condition vector features based on the flow matching model to obtain the robot's state residual trajectory, where the state residual trajectory represents the change of the state parameters of each trajectory point relative to the real-time state parameters.

[0074] For example, the flow matching model takes the adjusted conditional vector features as input, learns the evolution of the robot's state parameters in the continuous time domain, and predicts a smooth vector field pointing from the current real-time state parameters to the target state parameters. Based on this vector field, a continuous and abrupt state residual trajectory can be generated, making the robot's state changes conform to dynamic constraints and avoiding problems such as trajectory jumps, jitter, or discontinuities.

[0075] It can be understood that vector field prediction refers to the flow matching model learning the direction and amplitude distribution of the continuous change of state parameters over time, and outputting the state change increment at each moment, so that the trajectory maintains a smooth transition in terms of position, attitude and velocity, ensuring that subsequent joint control can stably track and will not cause the fuselage to become unstable due to sudden changes in trajectory.

[0076] Furthermore, the robot can determine its target planned trajectory based on the state residual trajectory and real-time state parameters. Specifically, the robot can obtain its initial motion trajectory within a preset time period based on the state residual trajectory and real-time state parameters, and then perform third-order spline interpolation on the initial motion trajectory to obtain the target planned trajectory. The third-order spline interpolation is used to make the density of the target planned trajectory greater than that of the initial motion trajectory.

[0077] For example, the robot can sequentially superimpose the increments of the centroid position, centroid orientation, centroid velocity, and joint angles at each moment in the state residual trajectory onto the current real-time state parameters to obtain an initial motion trajectory containing state information at multiple moments.

[0078] It can be understood that the state residual trajectory contains the state parameter increments corresponding to multiple time-series nodes within a preset time period. These state parameter increments characterize the magnitude of change in the robot's motion state parameters at each time-series node relative to the real-time state parameters. The real-time state parameters are the robot's uniquely determined motion state parameters at the current moment; a single real-time state parameter can only reflect the instantaneous working condition and cannot form a continuous motion path. The robot iteratively superimposes the state parameter increments corresponding to different time-series nodes onto the current real-time state parameters in chronological order, sequentially solving for the state parameters corresponding to each time-series node within the preset time period. The state parameters of all time-series nodes are integrated to form a continuous set of time-series state points. This set of state parameters corresponding to each time-series node, arranged in an orderly manner along the time dimension, characterizes the robot's pose change process, i.e., the robot's initial motion trajectory.

[0079] For example, the state residual trajectory represents the deviation between the target state parameter and the actual state parameter at each time node. The real-time state parameter pt at the current time t is the uniquely determined instantaneous state parameter. Using the condition vector ct=[pt,m_cmd,Et] as a condition, the flow matching model outputs the state residual trajectory Δp(t+τ) for each time node within a preset time period τ∈[0,T]. Δp(t+τ) is sequentially superimposed onto the real-time state parameter pt in chronological order to obtain the predicted state parameter ppred(t+τ) = pt + Δp(t+τ) at each time point. The set of temporal states composed of all predicted state parameters constitutes the initial motion trajectory. Based on this, the robot, combined with the real-time state parameters, performs third-order spline interpolation on the state residual trajectory to complete trajectory point densification and smoothing, thus obtaining the target planned trajectory.

[0080] It is understandable that, in response to problems such as large sampling intervals and abrupt curve transitions in the initial motion trajectory, third-order spline interpolation can significantly increase the sampling point density of the target planned trajectory, making the trajectory continuous and dense in the time dimension. This ensures the continuous and smooth transition of key motion parameters such as attitude, position, and angle, making the robot's motion more coherent and stable.

[0081] In some embodiments of this application, the target planned trajectory may include multiple sets of motion state parameters corresponding to each time node within a preset time period. The state parameters of the robot at a single time node may include the position, orientation, linear velocity, and angular velocity of the robot's center of mass, as well as the joint position, velocity, rotation angle, and angular velocity of each moving joint of the robot. The target planned trajectory may be formed by combining the motion state parameters of all time nodes in a time-ordered arrangement.

[0082] S104: Determine the target joint torque of the electronic device based on the target planned trajectory and real-time state parameters.

[0083] In some embodiments of this application, the robot can determine the target joint torque based on the target planned trajectory and real-time state parameters. It should be understood that joint torque is the physical quantity of power output that drives the rotation of each moving joint of the robot, adjusts its posture, and realizes displacement motion. It is a core execution parameter for controlling the overall motion posture and movement of the robot.

[0084] In some embodiments of this application, the target joint torque of the robot is determined based on the target planned trajectory and real-time state parameters. The specific process may include: first, compressing the target planned trajectory using an improved finite scalar quantization-based (iFSQ-based) motion tokenizer to obtain discrete target trajectory semantic tags. These target trajectory semantic tags are used to characterize the motion semantic information of each trajectory point in the target planned trajectory.

[0085] For example, a robot can extract the state parameter features of each temporal node in the target planned trajectory based on a motion segmenter, and define discrete semantic intervals through preset quantization rules, mapping continuous motion state data into standardized semantic units within a finite set, i.e., target trajectory semantic tags. For example, the target trajectory semantic tag corresponding to each trajectory point can be represented as discrete motion mode labels such as smooth straight movement, slight uphill climb, steering adjustment, and joint easing.

[0086] Then, the robot can combine historical state parameters to decode the semantic tags of the target trajectory to obtain the target joint position matrix, which is used to characterize the target state parameters of the robot's joints at each trajectory point.

[0087] For example, a robot can use an actor-critic architecture's action decoder, taking discrete target trajectory semantic tags and real-time state parameters as input. Relying on built-in semantic decoding mapping relationships and combined with time-accumulated historical state parameter constraints, it can reverse-parse the abstract discrete trajectory semantic tags into the joint target positions corresponding to each trajectory point, integrating them to form a target joint position matrix. For instance, the target joint position matrix is ​​a two-dimensional structured matrix, with rows and columns corresponding to joint numbers and time-series trajectory points, respectively. The joint dimension matches the number of the robot's mobile joints (e.g., when the robot has 29 mobile joints, the joint dimension of the target joint position matrix is ​​29). Each element in the target joint position matrix can specifically represent the target joint position at a single trajectory point, as well as the target joint angle, target joint velocity, or target joint angular velocity, etc.

[0088] S105: Based on the target joint torque, the electronic device is moved by the controller to move to the target position.

[0089] In some embodiments of this application, after the target joint moment matrix is ​​determined, the robot can control itself to move to the target position based on the target joint moment matrix through the controller.

[0090] For example, a robot can use a (high-frequency) proportional-differential controller, combined with a target joint position matrix and real-time state parameters, to calculate the deviation between the real-time joint position of each moving joint and the target joint position. Based on the deviation results, the joint torques corresponding to each joint are obtained, and these are combined to obtain the target joint torque matrix. The robot can then drive its moving joints to move using the target joint torque matrix based on the controller, thereby achieving smooth movement.

[0091] Based on the above method, regardless of the terrain the robot is in, it can dynamically generate a target planning trajectory adapted to the current terrain and calculate the corresponding target joint torque by combining terrain features, target motion commands, and the robot's real-time state parameters, thus achieving adaptive and stable control. For example, when the current terrain has undulations, potholes, or obstacles, the target planning trajectory can be corrected in a timely manner based on the real-time perceived terrain features and equipment status to calculate action commands such as stepping, obstacle avoidance, or restoring balance, preventing the equipment from becoming unbalanced and falling due to terrain incompatibility, and improving the environmental adaptability and safety of navigation control.

[0092] based on Figure 1 In addition to the motion control method shown, this application also provides a motion control system. Based on this motion control system, mobile electronic devices, such as mobile robots, can be controlled to implement the motion control method provided in the embodiments of this application.

[0093] For example, Figure 2A motion control system is shown, which may include a macro command module 201, a terrain and state condition generation module 202, and a high-frequency universal whole-body tracking module 203.

[0094] The macroscopic instruction module 201 is used to detect the target motion instruction of the electronic device (such as a mobile robot), and the target motion instruction corresponds to the target position; the terrain and state condition generation module 202 is used to detect the terrain features of the current position of the electronic device, and the terrain and state condition generation module 202 is used to generate the target planning trajectory of the electronic device based on the target motion instruction, terrain features and real-time state parameters of the electronic device; the high-frequency universal whole-body tracking module 203 is used to determine the target joint torque of the electronic device based on the target planning trajectory and real-time state parameters, and the high-frequency universal whole-body tracking module 203 is used to control the electronic device to move to the target position based on the target joint torque.

[0095] Thus, in this application, the functional modules work in a layered and coordinated manner, deeply integrating terrain features to complete dynamic trajectory correction and ensure stable high-frequency execution of actions by the equipment. Regardless of the terrain in which the electronic device is located, it can dynamically generate a target planning trajectory adapted to the terrain and calculate the target joint torque by combining terrain features, target motion commands, and real-time status parameters, thereby achieving full-domain adaptive stability control.

[0096] In some embodiments of this application, the terrain and state condition generation module 202 may include an environmental encoder, and the high-frequency general-purpose whole-body tracking module 203 may include a joint proportional differential controller; the environmental encoder may be used to detect the terrain features of the current location of the electronic device; the joint proportional differential controller may be used to determine the target joint torque of the electronic device, and control the movement of the electronic device based on the target joint torque to move to the target position.

[0097] Among them, the macroscopic instruction module 201 can operate based on a first frequency, the terrain and state condition generation module 202 can operate based on a second frequency, the high-frequency universal whole-body tracking module 203 can operate based on a third frequency, and the joint proportional differential controller can operate based on a fourth frequency. The fourth frequency is greater than the third frequency, the third frequency is greater than the second frequency, and the second frequency is greater than the first frequency.

[0098] In some embodiments of this application, the first frequency can be 1Hz to 5Hz, the second frequency can be 25Hz, the third frequency can be 50Hz, and the fourth frequency can be 200Hz. It is understood that the above frequency values ​​are merely illustrative, and this application does not specifically limit the values ​​of the first, second, third, and fourth frequencies.

[0099] Thus, the macroscopic instruction module can operate at low frequencies of 1Hz to 5Hz to meet the low real-time requirements of global navigation intent generation. The terrain and state condition generation module can operate at a mid-frequency of 25Hz to balance the complexity of terrain feature processing with the timeliness of trajectory updates. The high-frequency general-purpose whole-body tracking module can operate at a mid-to-high frequency of 50Hz to ensure the real-time performance of trajectory decoding and torque planning. The joint proportional differential controller can operate at a high frequency of 200Hz to achieve rapid response and closed-loop control of joint movements, with each level of frequency precisely matching the performance requirements of its corresponding function.

[0100] Furthermore, by entrusting the detection of complex terrain features to the environmental encoder operating at medium frequency, the joint proportional differential controller operating at high frequency does not need to participate in terrain data parsing, but only focuses on the rapid calculation and execution control of the target joint torque. This avoids the high-frequency module bearing redundant calculations, greatly saves system computing power consumption, and improves the stability and endurance of equipment operation.

[0101] In addition, the modules work collaboratively in a hierarchical manner according to frequency. The low-frequency module provides global command support, the mid-frequency module dynamically generates trajectories that adapt to the terrain, and the high-frequency module accurately executes the underlying actions and quickly corrects deviations, forming a closed-loop control chain of "global planning - dynamic adaptation - real-time execution". This not only ensures the flexibility and terrain adaptability of navigation control, but also ensures the accuracy of joint actions and the stability of equipment movement through high-frequency underlying execution, effectively improving the navigation control performance of electronic equipment in complex environments.

[0102] In some embodiments of this application, based on Figure 2 The control system shown Figure 3 A schematic diagram of the process of implementing the motion control method of this application based on the control system is shown.

[0103] refer to Figure 3 , Figure 3 The macro-level instruction layer (scene-objective navigator) in the middle can correspond to Figure 2 Macro instruction module 201 in the middle, Figure 3 The terrain-conditioned generative middleware in the middleware can correspond to... Figure 2 The terrain and state condition generation module 202 in the middle, Figure 3 The high-frequency general-purpose whole-body tracker in the text can correspond to... Figure 2 The high-frequency general-purpose whole-body tracking module 203 is included.

[0104] Continue to refer to Figure 3 The specific process of the motion control method may include the following.

[0105] First, the macroscopic instruction layer generates an ideal macroscopic velocity command based on the scene target and global attitude. This command, m_cmd = (Vx, Vy, Vyaw), includes forward velocity, lateral velocity, and yaw angle velocity. This macroscopic velocity command is then output to the terrain and state condition generation middleware. The scene target can be understood as corresponding to the target position image mentioned in S101 above, used to represent the semantic information of the target position. The global attitude can correspond to the pose (target position coordinates) of the target node in the world coordinate system mentioned in S101 above, used to represent the geometric and orientation information of the target position.

[0106] Secondly, the terrain and state condition generative middleware can acquire the real-time state parameters pt of the electronic device and generate terrain features Et through the built-in environment coding module. Then, based on the real-time state parameters pt, macroscopic motion commands m_cmd, and terrain features Et, a condition vector ct=[pt,m_cmd,Et] is formed. Furthermore, the generative model can predict the vector field of the condition vector based on the built-in flow matching model, output the state residual trajectory, and perform third-order spline interpolation on the state residual trajectory, i.e., trajectory point densification and smoothing, to obtain a continuous, high-density target planning trajectory, which is then output to the high-frequency general-purpose whole-body tracker.

[0107] Finally, the high-frequency universal whole-body tracker receives the target planned trajectory. First, it compresses the target planned trajectory using a motion segmenter based on improved finite scalar quantization to obtain the target planned trajectory semantic tag z_d. Then, it decodes the target planned trajectory semantic tag using an action decoder combined with real-time state parameters, including historical state parameters, to obtain the target joint position a_t. This target joint position can be represented in the form of a target joint position matrix, where the row and column dimensions correspond to the joint number and the time-series trajectory point, respectively. The (high-frequency) proportional-derivative controller can calculate the deviation based on the target joint position matrix and the real-time state parameters, outputting the target joint torque. This target joint torque can be represented in the form of a target joint torque matrix, and the controller drives the joints of the electronic device to perform actions based on the target joint torque matrix, completing the trajectory tracking control.

[0108] based on Figure 2 and Figure 3As can be seen, the navigation control architecture provided in this application is divided into three layers: macroscopic command generation, terrain and state condition generation, and high-frequency general-purpose whole-body tracking. The three layers operate hierarchically with low-frequency, medium-frequency, and high-frequency timing logic, respectively. The high-frequency working low-level whole-body control (WBC) unit can refer to a (high-frequency) proportional-derivative controller. It does not need to process and analyze complex terrain data, but only focuses on real-time joint torque calculation and motion closed-loop control, thereby effectively reducing the overall computing power overhead and ensuring the execution stability of the device's low-level joint movements.

[0109] Thus, the macro-level command layer can operate at low frequencies of 1Hz to 5Hz to meet the low real-time requirements of global navigation intent generation. The terrain and state condition generation module can operate at a mid-frequency of 25Hz to balance the complexity of terrain feature processing with the timeliness of trajectory updates. The high-frequency general-purpose whole-body tracker can operate at a mid-to-high frequency of 50Hz to ensure the real-time performance of trajectory decoding and torque planning. The joint proportional differential controller can operate at a high frequency of 200Hz to achieve rapid response and closed-loop control of joint movements, with each level of frequency precisely matching the performance requirements of its corresponding function.

[0110] Furthermore, in complex road conditions (such as slopes, uneven surfaces, gravel, and non-flat surfaces), the solution in this application introduces terrain feature parameters into the computational conditions of the flow matching model, enabling the trajectory generation process to identify environmental elements such as road surface undulations and terrain obstacles in advance. By combining environmental perception information with the motion characteristics of the equipment, adaptive actions such as balance adjustment, obstacle avoidance, and posture compensation can be generated autonomously.

[0111] Furthermore, the mobile control system provided in this application possesses hierarchical adaptive control capabilities. When the road surface is flat and the robot's posture is stable, the terrain and state condition generation layer reduces intervention, allowing the robot to maintain its initially planned target trajectory. When rugged terrain or posture instability is detected, the high-frequency general-purpose whole-body tracking layer adjusts the planned target trajectory and target joint torque output in real time, improving the robot's terrain adaptability and motion stability in unknown environments.

[0112] Furthermore, in some embodiments, this application also provides a readable storage medium storing instructions. When executed on an electronic device, the instructions cause the electronic device to implement the motion control method mentioned in this application.

[0113] Furthermore, in some embodiments, this application also provides a computer program product, including computer instructions. When the computer instructions are executed on an electronic device, they cause the electronic device to implement the motion control method mentioned in this application.

[0114] Furthermore, in some embodiments, this application also provides an electronic device. The electronic device includes at least one memory and at least one processor, with the memory coupled to the processor. The memory stores computer program code / instructions, which, when executed by the processor, enable the electronic device to implement the motion control method mentioned in this application.

[0115] For example, such as Figure 4 As shown, this application provides a schematic diagram of the structure of an electronic device. Figure 4 As shown, the electronic device 400 includes a processor 401, a communication interface 402, a memory 403, and a communication bus 404.

[0116] The memory 403 stores a computer program that can run on the processor 401. The memory 403 and the processor 401 communicate through the communication interface 402 and the communication bus 404.

[0117] Processor 401 may include general-purpose processors, including central processing units, neural network processors, etc.; it may also be digital signal processing (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0118] The memory 403 may include random access memory (RAM) or non-volatile memory, etc. Optionally, the memory 403 may also be at least one storage device located remotely from the processor 401.

[0119] The communication bus 404 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. This communication bus 404 can be divided into an address bus, a data bus, a control bus, etc.

[0120] It should be noted that the connection method between units / modules in the accompanying drawings provided in this application is only one example. In other embodiments, units / modules can be directly connected via buses, signal lines, or connectors, or they can be indirectly connected.

[0121] Furthermore, some structural or methodological features may be shown in a specific arrangement and / or order in the accompanying drawings provided in this application. However, it should be understood that such a specific arrangement and / or order may not be necessary. For example, in other embodiments, these features may be arranged in a manner and / or order different from that shown in the illustrative drawings. Additionally, including structural or methodological features in a particular figure does not imply that such features are required in all embodiments; for example, in other embodiments, these features may be omitted or may be combined with other features.

[0122] It should be noted that all units / modules mentioned in the device embodiments of this application are logical units / modules. Physically, a logical unit / module can be a physical unit / module, a part of a physical unit / module, or a combination of multiple physical units / modules. The physical implementation of these logical units / modules themselves is not the most important factor; the combination of functions implemented by these logical units / modules is the key to solving the technical problems proposed in this application. Furthermore, to highlight the innovative aspects of this application, the above-described device embodiments of this application have not introduced units / modules that are not closely related to solving the technical problems proposed in this application. This does not mean that the above-described device embodiments do not contain other units / modules.

[0123] The embodiments disclosed in this application can be implemented in hardware, software, firmware, or a combination of these implementation methods. Embodiments of this application can be implemented as computer programs or program code executable on a programmable system, the programmable system including at least one processor, a storage system (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device.

[0124] In some cases, the disclosed embodiments may be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also be implemented as instructions carried on or stored thereon by one or more transient or non-transitory machine-readable (e.g., computer-readable) storage media, which may be read and executed by one or more processors.

[0125] It should be noted that in the examples and description of this application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0126] Although this application has been illustrated and described with reference to certain preferred embodiments thereof, those skilled in the art will understand that various changes in form and detail may be made thereto without departing from the scope of this application.

Claims

1. A motion control method, characterized in that, Applied to a portable electronic device, the method includes: A target motion command is detected to instruct the electronic device to move, the target motion command corresponding to a target position; Detect the terrain features of the current location of the electronic device; Based on the target motion command, the terrain features, and the real-time status parameters of the electronic device, the target planned trajectory of the electronic device is generated; The target joint torque of the electronic device is determined based on the target planned trajectory and the real-time state parameters. Based on the target joint torque, the electronic device is moved by the controller to the target position.

2. The motion control method according to claim 1, characterized in that, The target motion command includes the longitudinal velocity, lateral velocity, and yaw rate of the electronic device; The target motion command is determined based on navigation target information, wherein the navigation target information includes a target position image and / or target position coordinates, the target position image is used to characterize the semantic information of the target position, and the target position coordinates are used to characterize the geometric information of the target position.

3. The motion control method according to claim 2, characterized in that, The methods for determining the terrain features include: The geometric features of the environment in which the electronic device is located are obtained, and the geometric features are input into a first neural network for feature compression to obtain the terrain features in the form of latent variables.

4. The motion control method according to claim 3, characterized in that, The electronic device includes multiple movable joints; The real-time status parameters of the electronic device include at least one of the following: the position, orientation, linear velocity, and angular velocity of the center of mass of the electronic device, and the joint position, velocity, rotation angle, and angular velocity of each of the moving joints.

5. The motion control method according to claim 4, characterized in that, The step of generating the target planning trajectory of the electronic device based on the target motion command, the terrain features, and the real-time status parameters of the electronic device includes: The target motion command, the terrain features, and the real-time state parameters are concatenated to obtain a condition vector; The time step information and the conditional vector are processed based on the normalization mechanism to generate scaling parameters and offset parameters for feature extraction, and the scaling parameters and offset parameters are input into the second neural network. Wherein, the time step information represents the current time point, and the scaling parameter and the offset parameter are used to adjust the conditional vector features extracted by the second neural network to obtain the adjusted conditional vector features; Based on the flow matching model, the adjusted conditional vector features are used to predict the vector field to obtain the state residual trajectory of the electronic device, wherein the state residual trajectory represents the change of the state parameters of each trajectory point relative to the real-time state parameters. The target planning trajectory of the electronic device is determined based on the state residual trajectory and the real-time state parameters.

6. The motion control method according to claim 5, characterized in that, The step of concatenating the target motion command, the terrain features, and the real-time state parameters to obtain a condition vector includes: If the terrain feature indicates that the location of the electronic device is on a non-flat road surface, then the target motion command is adjusted based on the terrain feature, and the adjusted target motion command, the terrain feature, and the real-time state parameters are concatenated to obtain the condition vector.

7. The motion control method according to claim 5, characterized in that, Determining the target planning trajectory of the electronic device based on the state residual trajectory and the real-time state parameters includes: Based on the state residual trajectory and the real-time state parameters, the initial motion trajectory of the electronic device within a preset time period is obtained. The initial motion trajectory is then subjected to third-order spline interpolation to obtain the target planned trajectory. The third-order spline interpolation is used to make the density of the target planned trajectory greater than that of the initial motion trajectory.

8. The motion control method according to claim 7, characterized in that, Determining the target joint torque of the electronic device based on the target planned trajectory and the real-time state parameters includes: The target planning trajectory is compressed using a motion segmenter based on finite scalar quantization to obtain discrete target trajectory semantic tags, wherein the target trajectory semantic tags are used to characterize the motion semantic information of each trajectory point in the target planning trajectory; By combining the historical state parameters of the electronic device, the semantic markers of the target trajectory are decoded to obtain the target joint position matrix, wherein the target joint position matrix is ​​used to characterize the target state parameters of the joints of the electronic device at each trajectory point; Based on the target joint position matrix and the real-time state parameters, the target joint torque matrix of the electronic device is determined.

9. A mobile control system, characterized in that, The mobility control system includes a macro-command module, a terrain and state condition generation module, and a high-frequency universal whole-body tracking module. The macroscopic instruction module is used to detect target motion instructions of the electronic device, and the target motion instructions correspond to the target position; The terrain and state condition generation module is used to detect the terrain features of the current location of the electronic device, and the terrain and state condition generation module is used to generate the target planning trajectory of the electronic device based on the target motion command, the terrain features and the real-time state parameters of the electronic device. The high-frequency universal whole-body tracking module is used to determine the target joint torque of the electronic device based on the target planned trajectory and the real-time state parameters, and the high-frequency universal whole-body tracking module is used to control the electronic device to move to the target position based on the target joint torque.

10. The mobile control system according to claim 9, characterized in that, The terrain and state condition generation module includes an environmental encoder, and the high-frequency universal whole-body tracking module includes a joint proportional differential controller. The environmental encoder is used to detect the terrain features of the current location of the electronic device; The joint proportional differential controller is used to determine the target joint torque of the electronic device, and control the movement of the electronic device based on the target joint torque to move to the target position; Among them, the macro instruction module operates based on a first frequency, the terrain and state condition generation module operates based on a second frequency, the high-frequency universal whole-body tracking module operates based on a third frequency, and the joint proportional differential controller operates based on a fourth frequency; The fourth frequency is greater than the third frequency, the third frequency is greater than the second frequency, and the second frequency is greater than the first frequency.

11. An electronic device, characterized in that, include: At least one memory and at least one processor, the memory being coupled to the processor, the memory being used to store computer program code / instructions; When the computer program code / instructions are executed by the processor, the electronic device implements the motion control method as described in any one of claims 1 to 8.

12. A readable storage medium, characterized in that, The readable storage medium stores instructions that, when executed on an electronic device, cause the electronic device to implement the motion control method as described in any one of claims 1 to 8.

13. A computer program product, characterized in that, include: Computer instructions, when executed on an electronic device, cause the electronic device to perform the motion control method as described in any one of claims 1 to 8.

Images