Robot motion control strategy determination method and device and electronic equipment

Abstract

The invention discloses a method and a device for determining a motion control strategy of a robot and electronic equipment. The method comprises the following steps: acquiring an obstacle space parameter, an obstacle hardness parameter and a joint motion parameter; determining a plurality of candidate path areas according to the obstacle space parameters and the obstacle hardness parameters; determining motion risk parameters respectively corresponding to the plurality of candidate path areas; determining stability constraint parameters corresponding to the plurality of candidate path areas according to the joint motion parameters and the obstacle hardness parameters; determining a target path area according to the motion risk parameters and the stability constraint parameters corresponding to the plurality of candidate path areas; and determining a motion control strategy according to the motion risk parameters of the target path area. The technical problem that in the related technology, when motion control is conducted on the robot, a determined motion control strategy is difficult to adapt to a complex unstructured environment, and then the stability and reliability of robot motion are insufficient is solved.

Description

Technical Field

[0001] This invention relates to the field of robotics, and more specifically, to a method, apparatus, and electronic device for determining motion control strategies for robots. Background Technology

[0002] In related technologies, to ensure that robots can complete motion operations according to task requirements while maintaining posture stability during movement, avoiding collisions with obstacles in the environment, and ensuring task execution efficiency and robot safety, it is necessary to determine a reasonable motion control strategy. However, in related technologies, when controlling the motion of robots, there is a technical problem that a fixed motion control strategy is difficult to adapt to complex unstructured environments, leading to insufficient stability and reliability of robot motion.

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

[0004] This invention provides a method, apparatus, and electronic device for determining motion control strategies for robots, in order to at least solve the technical problem in the related art where a determined motion control strategy is difficult to adapt to complex unstructured environments, resulting in insufficient stability and reliability of robot motion.

[0005] According to one aspect of the present invention, a method for determining a robot's motion control strategy is provided, comprising: acquiring obstacle space parameters, obstacle stiffness parameters, and joint motion parameters of the robot; determining a plurality of candidate path regions of the robot based on the obstacle space parameters and the obstacle stiffness parameters; determining motion risk parameters corresponding to the plurality of candidate path regions respectively, wherein the corresponding motion risk parameters represent the degree of risk of the robot's movement in the corresponding candidate path region; determining stability constraint parameters corresponding to the plurality of candidate path regions respectively based on the joint motion parameters and the obstacle stiffness parameters, wherein the corresponding stability constraint parameters represent the degree of constraint of the corresponding candidate path region on the stability of the robot's motion posture; determining a target path region from the plurality of candidate path regions based on the motion risk parameters and stability constraint parameters corresponding to the plurality of candidate path regions respectively; and determining a motion control strategy of the robot based on the motion risk parameters of the target path region.

[0006] Optionally, based on the joint motion parameters and the obstacle hardness parameters, determining stability constraint parameters corresponding to the plurality of candidate path regions includes: determining joint angle parameters, joint load parameters, and joint state parameters corresponding to the robot based on the joint motion parameters; determining motion posture feature parameters corresponding to the robot based on the joint angle parameters; determining obstacle hardness categories corresponding to the plurality of candidate path regions based on the obstacle hardness parameters; determining initial constraint parameters corresponding to the plurality of candidate path regions based on the motion posture feature parameters and the obstacle hardness categories corresponding to the plurality of candidate path regions; and correcting the initial constraint parameters corresponding to the plurality of candidate path regions based on the joint load parameters and the joint state parameters to obtain stability constraint parameters corresponding to the plurality of candidate path regions.

[0007] Optionally, determining the motion risk parameters corresponding to the plurality of candidate path regions includes: determining the robot's historical collision data; determining collision damage indices corresponding to multiple obstacle hardness categories based on the historical collision data, wherein the obstacle hardness parameters include multiple obstacle hardness categories; determining obstacle distribution characteristics corresponding to the multiple obstacle hardness categories for any target candidate path region among the plurality of candidate path regions; determining motion risk parameters corresponding to the target candidate path region based on the collision damage indices and obstacle distribution characteristics corresponding to the multiple obstacle hardness categories; and determining motion risk parameters corresponding to other candidate path regions besides the target candidate path region among the plurality of candidate path regions by using the same method as determining motion risk parameters corresponding to the target candidate path region.

[0008] Optionally, before obtaining the obstacle space parameters of the robot, the method further includes: determining a visual image of the robot; performing a masking operation on multiple image pixels of the visual image to obtain a mask image corresponding to the visual image, wherein the masking operation is used to perform feature enhancement processing on a first pixel among the multiple image pixels and feature reduction processing on a second pixel among the multiple image pixels, the first pixel being the pixel corresponding to an obstacle among the multiple image pixels, and the second pixel being the pixel representing a non-obstacle among the multiple image pixels; determining contour parameters corresponding to the obstacle based on the mask image, wherein the contour parameters include multiple contour pixels and position parameters corresponding to the multiple contour pixels respectively; determining depth parameters corresponding to the multiple contour pixels respectively based on the visual image; and determining obstacle space parameters corresponding to the robot based on the contour parameters and the depth parameters corresponding to the multiple contour pixels respectively.

[0009] Optionally, determining multiple candidate path regions for the robot based on the obstacle space parameters and the obstacle hardness parameters includes: determining the robot's action space, wherein the action space includes multiple spatial units, and the action space is the three-dimensional space corresponding to the robot's target action; determining obstacle features corresponding to the multiple spatial units based on the obstacle space parameters and the obstacle hardness parameters, wherein the corresponding obstacle features represent the characteristics of the obstacles in the corresponding spatial units; and determining multiple candidate path regions for the robot based on the obstacle features corresponding to the multiple spatial units.

[0010] Optionally, determining multiple candidate path regions of the robot based on the obstacle features corresponding to the multiple spatial units includes: determining adjacent units corresponding to the multiple spatial units; for any target spatial unit among the multiple spatial units, determining a similarity index between the target spatial unit and its adjacent units based on the obstacle features of the target spatial unit and the obstacle features of its adjacent units; determining a result of whether to perform a merging operation corresponding to the target spatial unit based on the similarity index, wherein the merging operation is used to merge the target spatial unit with its adjacent units; determining a result of whether to perform a merging operation corresponding to the target spatial unit by using the same method as determining the result of whether to perform a merging operation corresponding to the target spatial unit; and determining multiple candidate path regions of the robot based on the results of whether to perform a merging operation corresponding to the multiple spatial units.

[0011] Optionally, before obtaining the obstacle hardness parameters of the robot, the method further includes: determining the force data of the robot within a target time period; determining the force change parameters corresponding to multiple sub-time periods based on the force data, wherein the target time period includes multiple sub-time periods; determining the force intensity parameters corresponding to the robot based on the force change parameters corresponding to the multiple sub-time periods; and determining the obstacle hardness parameters corresponding to the robot based on the force intensity parameters.

[0012] According to one aspect of the present invention, a motion control strategy determination device for a robot is provided, comprising: an acquisition module for acquiring obstacle space parameters, obstacle stiffness parameters, and joint motion parameters of the robot; a first determination module for determining a plurality of candidate path regions of the robot based on the obstacle space parameters and the obstacle stiffness parameters; a second determination module for determining motion risk parameters corresponding to the plurality of candidate path regions, wherein the corresponding motion risk parameters represent the degree of risk of the robot moving in the corresponding candidate path region; a third determination module for determining stability constraint parameters corresponding to the plurality of candidate path regions based on the joint motion parameters and the obstacle stiffness parameters, wherein the corresponding stability constraint parameters represent the degree of constraint of the corresponding candidate path region on the motion posture stability of the robot; a fourth determination module for determining a target path region from the plurality of candidate path regions based on the motion risk parameters and stability constraint parameters corresponding to the plurality of candidate path regions; and a fifth determination module for determining a motion control strategy for the robot based on the motion risk parameters of the target path region.

[0013] According to one aspect of the present invention, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the motion control strategy determination method for a robot as described in any of the preceding embodiments.

[0014] According to one aspect of the present invention, a computer-readable storage medium is provided, wherein when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the motion control strategy determination method for a robot as described above.

[0015] In this embodiment of the invention, obstacle space parameters, obstacle stiffness parameters, and joint motion parameters of the robot are obtained; based on the obstacle space parameters and obstacle stiffness parameters, multiple candidate path regions of the robot are determined; motion risk parameters corresponding to each of the multiple candidate path regions are determined, wherein the corresponding motion risk parameters represent the degree of risk of the robot's movement in the corresponding candidate path region; based on the joint motion parameters and obstacle stiffness parameters, stability constraint parameters corresponding to each of the multiple candidate path regions are determined, wherein the corresponding stability constraint parameters are used to represent the degree of constraint of the corresponding candidate path region on the stability of the robot's motion posture; based on the motion risk parameters and stability constraint parameters corresponding to each of the multiple candidate path regions, a target path region is determined from the multiple candidate path regions; based on the motion risk parameters of the target path region, the robot's motion control strategy is determined. By acquiring the spatial and hardness parameters of obstacles, the robot can comprehensively perceive obstacle information in the environment. By comprehensively considering the spatial and hardness parameters of obstacles, multiple candidate path regions are divided, providing the robot with a variety of choices. Based on this, by evaluating the risk level of each candidate path region and the degree of constraint of each candidate path region on the stability of the robot's motion posture, the target path region is selected from multiple candidate path regions. This ensures that the selected path region is both safe and stable, thereby solving the technical problem in related technologies where a fixed motion control strategy is difficult to adapt to complex unstructured environments, resulting in insufficient stability and reliability of robot motion. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0017] Figure 1 This is a flowchart of a method for determining the motion control strategy of a robot according to an embodiment of the present invention;

[0018] Figure 2 This is a flowchart of a method for determining the motion control strategy of a robot in an optional embodiment of the present invention;

[0019] Figure 3 This is a schematic diagram of the motion control system structure in an optional embodiment of the present invention;

[0020] Figure 4 This is a structural block diagram of a robot motion control strategy determination device according to an embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] First, some nouns or terms that appear in the description of the embodiments of this application shall be interpreted as follows:

[0024] DeepLab V3+: DeepLab V3+ is a deep learning model for image segmentation, specifically for semantic segmentation tasks.

[0025] ResNet101: ResNet101 is a deep residual network that contains multiple residual blocks, each of which consists of multiple convolutional layers.

[0026] Pascal VOC: Pascal VOC is a benchmark dataset for object detection and segmentation tasks in the field of computer vision.

[0027] Canny edge detection algorithm: Canny edge detection is a classic edge detection algorithm used to extract edges from images.

[0028] Actor Network: An Actor Network is a policy network that is responsible for selecting actions based on the current state.

[0029] Critic Network: A Critic network is a value function network responsible for evaluating the quality of the actions chosen by the Actor.

[0030] ReLU activation function: The ReLU (Rectified Linear Unit) activation function is an activation function used to alleviate the gradient vanishing problem.

[0031] tanh activation function: The tanh (Hyperbolic Tangent) activation function is an activation function that maps the input to a symmetric interval.

[0032] Q-value: The Q-value represents the expected reward for taking a certain action in a given state.

[0033] Savitzky-Golay filtering: Savitzky-Golay filtering is a smoothing filtering method based on polynomial fitting.

[0034] Example 1

[0035] According to an embodiment of the present invention, an embodiment of a method for determining a robot's motion control strategy is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0036] Figure 1 This is a flowchart of a robot motion control strategy determination method according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0037] S102, Obtain the robot's obstacle space parameters, obstacle hardness parameters, and joint motion parameters;

[0038] In step S102 of this application, the obstacle space parameters, obstacle hardness parameters, and joint motion parameters of the robot are obtained.

[0039] This involves robots, which are mechanical devices capable of autonomous movement.

[0040] This involves obstacle spatial parameters, which are parameters used to describe the position and distribution of obstacles in space, such as the coordinate position of the obstacle in three-dimensional space, the distribution of the obstacle in the robot's motion path, the size information of the obstacle, and the shape information of the obstacle.

[0041] This involves obstacle hardness parameters, which are parameters used to describe the hardness characteristics of obstacles and include obstacle hardness categories.

[0042] This involves joint motion parameters, which are parameters related to the robot's joint motion and are used to reflect the robot's joint motion characteristics.

[0043] The purpose of acquiring obstacle space parameters, obstacle hardness parameters, and joint motion parameters of the robot is to comprehensively perceive obstacle information (including position, distribution, hardness, etc.) in the robot's environment and the robot's own motion state, thereby providing a data foundation for selecting the optimal motion path and improving the robot's motion stability and reliability in complex environments.

[0044] S104, Based on obstacle space parameters and obstacle hardness parameters, determine multiple candidate path regions for the robot;

[0045] In step S104 provided in this application, multiple candidate path regions for the robot are determined based on obstacle space parameters and obstacle hardness parameters.

[0046] This involves multiple candidate path regions, which are several alternative areas selected by the robot when planning its movement path based on obstacle spatial parameters and obstacle hardness parameters. These areas are feasible for movement and form the basis for further selection of target path regions.

[0047] By comprehensively considering the spatial location, distribution, and hardness characteristics of obstacles, robots can be provided with diverse path options in complex environments, reducing collision risks, optimizing path planning, and thus improving the robot's motion stability and reliability in complex environments.

[0048] S106, determine the motion risk parameters corresponding to the multiple candidate path areas respectively, wherein the corresponding motion risk parameters represent the degree of risk of the robot moving in the corresponding candidate path area;

[0049] In step S106 of this application, motion risk parameters corresponding to multiple candidate path regions are determined.

[0050] This includes motion risk parameters, which are used to quantify the level of risk that a robot may face when moving in each candidate path area. These motion risk parameters are used to assess the level of risk that the robot faces when moving in the area, such as collision damage and loss of posture, based on the quantification of obstacle characteristics and environmental features of the candidate path area.

[0051] Motion risk parameters are used to quantify the degree of risk that a robot may face when moving in each candidate path region. By quantifying the risk level of candidate path regions through motion risk parameters, the robot can intuitively compare the risk levels of different path regions, thereby making a better path selection, effectively reducing the risk of collision and attitude loss during movement, and improving the robot's motion stability and reliability in complex environments.

[0052] Furthermore, this motion risk parameter takes into account the following factors:

[0053] The distribution characteristics of obstacles include their density, location, shape, and size within the path area. For example, the denser the obstacle distribution, the higher the movement risk within the path area.

[0054] Obstacle hardness characteristics: The hardness category of an obstacle (e.g., hard, medium, soft) affects the degree of injury upon collision. Hard obstacles may result in greater collision damage, thus increasing the risk of injury during movement.

[0055] Historical collision data: By analyzing the robot's historical collision records in similar environments, collision damage indices for obstacles of different hardness are calculated. This data is used to assess the risk level of the current path area.

[0056] Path complexity: The complexity of the path within the path region, such as the degree of tortuosity and narrowness of the path, also affects the motion risk. Complex paths may make the robot more prone to collisions or posture instability.

[0057] S108. Based on the joint motion parameters and obstacle stiffness parameters, determine the stability constraint parameters corresponding to the multiple candidate path regions respectively. The corresponding stability constraint parameters are used to represent the degree of constraint of the corresponding candidate path region on the stability of the robot's motion posture.

[0058] In step S108 provided in this application, stability constraint parameters corresponding to multiple candidate path regions are determined based on joint motion parameters and obstacle stiffness parameters.

[0059] This involves stability constraint parameters, which are used to quantify the degree of constraint imposed by the candidate path region on the stability of the robot's motion posture, and reflect the robot's ability to maintain stable motion in the candidate path region.

[0060] Stability constraint parameters are used to quantify the stability of a robot's motion posture when moving in different candidate path regions. By determining the stability constraint parameters, the robot can more accurately assess the degree of constraint on its motion posture stability by each candidate path region, avoid motion imbalance caused by ignoring posture constraints, and thus improve the stability and reliability of the robot's motion in complex unstructured environments.

[0061] S110, Based on the motion risk parameters and stability constraint parameters corresponding to multiple candidate path regions, the target path region is determined from the multiple candidate path regions;

[0062] In step S110 of this application, the target path region is determined from the multiple candidate path regions based on the motion risk parameters and stability constraint parameters corresponding to the multiple candidate path regions.

[0063] This involves the target path region, which is the optimal path region selected from multiple candidate path regions by comprehensively evaluating the motion risk parameters and stability constraint parameters of each candidate path region.

[0064] The selection of the target path region is based on the following criteria:

[0065] Lowest risk of movement: The robot faces the lowest level of risk when moving within this path area.

[0066] Optimal stability: The robot can maintain the highest posture stability when moving within this path area, and is minimally affected by factors such as the hardness and distribution of obstacles.

[0067] The optimal overall assessment is that the target path region is the best choice after comprehensively considering motion risks and stability constraints, balancing safety (low risk) and reliability (high stability).

[0068] The target path region is the optimal path area selected from multiple candidate path regions based on a comprehensive evaluation of the motion risk parameters and stability constraint parameters of each candidate path region. It represents the path with the lowest risk and best stability for the robot under the current environmental conditions, significantly improving the robot's motion safety and reliability in complex environments. Determining the target path region enables the robot to intelligently select the optimal path based on obstacle information in the environment and its own motion state, thereby achieving stable and reliable motion in complex unstructured environments and significantly improving the success rate and efficiency of task execution.

[0069] S112, determine the robot's motion control strategy based on the motion risk parameters of the target path area.

[0070] In step S112 of this application, the motion control strategy of the robot is determined based on the motion risk parameters of the target path area.

[0071] This involves a motion control strategy, which is a specific motion plan and control strategy formulated for the robot based on the motion risk parameters of the target path area, and is used to control the robot to safely complete the motion task within the selected target path area.

[0072] The formulation of motion control strategies enables robots to dynamically adjust their motion state based on motion risk parameters within the target path area, thereby achieving safe and stable motion in complex environments. This strategy not only improves the robot's motion safety but also optimizes task execution efficiency.

[0073] Furthermore, this motion control strategy includes the following aspects:

[0074] Speed ​​planning: Adjust the robot's speed based on motion risk parameters of the target path area. For example, in areas approaching hard obstacles, the robot may slow down to reduce the impact force upon collision.

[0075] Trajectory Adjustment: Based on the distribution and hardness of obstacles within the target path area, the robot's trajectory is adjusted to select a safer path and avoid high-risk areas. For example, when there are hard obstacles within the path area, the robot may choose to detour rather than pass directly through them.

[0076] Joint motion control: Adjusting the robot's joint motion based on stability constraint parameters of the target path region. For example, adjusting joint angles, loads, and motion states to maintain the robot's posture stability during movement.

[0077] Real-time feedback and adjustment: By monitoring the robot's motion state and environmental changes in real time, the motion control strategy is dynamically adjusted. For example, if a new obstacle or environmental change is detected during movement, the robot can adjust its path or speed in real time.

[0078] Through the above steps S102-S110, the obstacle space parameters, obstacle stiffness parameters, and joint motion parameters of the robot are obtained; based on the obstacle space parameters and obstacle stiffness parameters, multiple candidate path regions of the robot are determined; motion risk parameters corresponding to each of the multiple candidate path regions are determined, where the corresponding motion risk parameter represents the degree of risk of the robot's movement in the corresponding candidate path region; based on the joint motion parameters and obstacle stiffness parameters, stability constraint parameters corresponding to each of the multiple candidate path regions are determined, where the corresponding stability constraint parameter is used to represent the degree of constraint of the corresponding candidate path region on the stability of the robot's motion posture; based on the motion risk parameters and stability constraint parameters corresponding to each of the multiple candidate path regions, the target path region is determined from the multiple candidate path regions; based on the motion risk parameters of the target path region, the robot's motion control strategy is determined. By acquiring the spatial and hardness parameters of obstacles, the robot can comprehensively perceive obstacle information in the environment. By comprehensively considering the spatial and hardness parameters of obstacles, multiple candidate path regions are divided, providing the robot with a variety of choices. Based on this, by evaluating the risk level of each candidate path region and the degree of constraint of each candidate path region on the stability of the robot's motion posture, the target path region is selected from multiple candidate path regions. This ensures that the selected path region is both safe and stable, thereby solving the technical problem in related technologies where a fixed motion control strategy is difficult to adapt to complex unstructured environments, resulting in insufficient stability and reliability of robot motion.

[0079] As an optional embodiment, stability constraint parameters corresponding to multiple candidate path regions are determined based on joint motion parameters and obstacle hardness parameters. This includes: determining joint angle parameters, joint load parameters, and joint state parameters corresponding to the robot based on joint motion parameters; determining motion posture feature parameters corresponding to the robot based on joint angle parameters; determining obstacle hardness categories corresponding to multiple candidate path regions based on obstacle hardness parameters; determining initial constraint parameters corresponding to multiple candidate path regions based on motion posture feature parameters and obstacle hardness categories corresponding to multiple candidate path regions; and correcting the initial constraint parameters corresponding to multiple candidate path regions based on joint load parameters and joint state parameters to obtain stability constraint parameters corresponding to multiple candidate path regions.

[0080] In this embodiment, the specific steps for determining the stability constraint parameters corresponding to multiple candidate path regions are described based on joint motion parameters and obstacle stiffness parameters.

[0081] This involves joint angle parameters, which are used to reflect the current rotation angle of each joint of the robot. By using joint angle parameters, the current posture and position of the robot can be determined, and its motion posture stability and flexibility can be evaluated.

[0082] This includes joint load parameters, which reflect the joint load (force or torque) experienced by each joint of the robot during movement. These parameters can be used to assess the magnitude of the load on a joint during movement; excessive load may lead to joint instability or damage.

[0083] This includes joint state parameters, which reflect the real-time operating status of each joint of the robot and are used to indicate whether the joint is in a normal working state.

[0084] This involves motion posture characteristic parameters, which are parameters used to reflect the overall motion posture characteristics of the robot, such as the robot's overall tilt angle and posture change rate.

[0085] This involves the classification of obstacle hardness, which refers to the category to which the hardness of an obstacle belongs. This classification can be based on hardness levels; for example, obstacle 1 is classified as "hard" (e.g., rock), obstacle 2 as "medium hard" (e.g., wood), and obstacle 3 as "soft" (e.g., sponge). Alternatively, it can be based on physical material, including rock (high hardness, high risk of collision damage), wood (medium hardness), and plastic (low hardness, low risk of collision damage), etc.

[0086] This involves initial constraint parameters, which are parameters initially determined based on motion posture feature parameters and obstacle hardness categories, and are used to constrain the robot's motion posture in the candidate path area.

[0087] By determining joint angles, loads, and state parameters based on joint motion parameters, and obstacle hardness categories based on obstacle hardness parameters, initial constraint parameters are determined and corrected to obtain stability constraint parameters. This approach comprehensively considers the robot's own motion capabilities and fully integrates the environmental characteristics of the path area. It enables a comprehensive evaluation of the robot's motion stability and safety in different candidate path areas, which not only improves the scientific nature and accuracy of path planning but also enhances the robot's adaptability and task execution success rate in complex environments, effectively reduces collision risks, and ensures reliable and safe motion.

[0088] As an optional embodiment, determining motion risk parameters corresponding to multiple candidate path regions includes: determining the robot's historical collision data; determining collision damage indices corresponding to multiple obstacle hardness categories based on the historical collision data, wherein the obstacle hardness parameters include multiple obstacle hardness categories; determining obstacle distribution characteristics corresponding to multiple obstacle hardness categories for any target candidate path region among the multiple candidate path regions; determining motion risk parameters corresponding to the target candidate path region based on the collision damage indices and obstacle distribution characteristics corresponding to multiple obstacle hardness categories; and determining motion risk parameters corresponding to other candidate path regions besides the target candidate path region by using the method of determining motion risk parameters corresponding to the target candidate path region.

[0089] This embodiment describes the specific steps for determining motion risk parameters corresponding to multiple candidate path regions.

[0090] This includes historical collision data, which is the data recorded when the robot collided with obstacles of different hardness categories during its past movements.

[0091] This includes a collision damage index, which is used to quantify the degree of damage caused to the robot by obstacles of different hardness categories based on historical collision data.

[0092] This involves the selection of a target candidate path area, which is the currently selected candidate path area from multiple candidate path areas used to calculate motion risk parameters. In this process, the risk parameters of one candidate path area can be calculated first, and then this calculation method can be replicated to all other candidate path areas, ultimately achieving batch determination of risk parameters for all areas.

[0093] This involves obstacle distribution characteristics, which are used to reflect the spatial distribution characteristics of obstacles of different hardness categories within the target candidate path area.

[0094] For any candidate path region, by determining the distribution characteristics of each obstacle hardness category, the spatial distribution of obstacles with different levels of hazard in that path region can be accurately understood, providing environmental details for subsequent risk calculation. By combining the collision damage index (quantifying the degree of obstacle hazard) of each hardness category with the distribution characteristics to determine the motion risk parameters of the target path region, a comprehensive consideration of the magnitude of obstacle hazard and collision probability can be achieved, thus realizing the precise quantification of path region risk.

[0095] As an optional embodiment, before obtaining the obstacle space parameters of the robot, the method further includes: determining a visual image of the robot; performing a masking operation on multiple image pixels of the visual image to obtain a mask image corresponding to the visual image, wherein the masking operation is used to perform feature enhancement processing on a first pixel among the multiple image pixels and feature reduction processing on a second pixel among the multiple image pixels, the first pixel being the pixel corresponding to the obstacle among the multiple image pixels, and the second pixel being the pixel representing the pixel corresponding to a non-obstacle among the multiple image pixels; determining contour parameters corresponding to the obstacle based on the mask image, wherein the contour parameters include multiple contour pixels and position parameters corresponding to the multiple contour pixels respectively; determining depth parameters corresponding to the multiple contour pixels respectively based on the visual image; and determining obstacle space parameters corresponding to the robot based on the contour parameters and the depth parameters corresponding to the multiple contour pixels respectively.

[0096] This embodiment describes the specific steps before obtaining the robot's obstacle space parameters.

[0097] This involves visual images, which are images that reflect information about the robot's surrounding environment.

[0098] This involves multiple image pixels, which are pixels in a visual image.

[0099] This involves masking operations, an image processing technique that classifies and enhances or weakens pixels in a visual image. The core objective is to highlight obstacle areas and suppress background interference. Specifically, the operation involves enhancing the features of pixels corresponding to obstacles (first pixels) (e.g., increasing contrast and strengthening pixel values), and weakening the features of pixels corresponding to non-obstacles (second pixels) (e.g., reducing brightness and weakening pixel values). This operation can be implemented using semantic segmentation algorithms, such as DeepLab V3+.

[0100] This involves a mask image, which is an image generated after a masking operation. The mask image can be a binary mask image, for example, the first pixel corresponding to an obstacle is marked as 1, and the second pixel corresponding to a non-obstacle is marked as 0, so that obstacles and background areas in the image can be clearly distinguished, laying the foundation for subsequent extraction of obstacle contours.

[0101] This involves the first pixel, which is the pixel in the visual image that directly corresponds to the obstacle.

[0102] This involves the second pixel, which is the pixel in the visual image that corresponds to the non-obstacle.

[0103] This involves obstacles, which are objects in the robot's movement path that may hinder its normal movement and pose a collision risk, such as rocks, trees, boxes, and steps. These objects form independent visual regions in the visual image, which the robot needs to identify and avoid.

[0104] This includes non-obstacles, which are objects or areas in the robot's movement path that do not impede its movement and pose no risk of collision.

[0105] This involves contour parameters, which are used to describe the shape and contour of an obstacle. For example, the contour parameters of a rectangular obstacle would include all the pixels on its four sides. These pixels, arranged in order, can completely represent the two-dimensional boundary shape of the obstacle.

[0106] This involves multiple contour pixels, which are pixels corresponding to the contour of the obstacle. They can be understood as pixels on the contour of the obstacle.

[0107] This involves positional parameters, which are used to represent the position of the corresponding contour pixels. These parameters can be used to determine the specific location of the contour pixels in the image and can be represented by coordinates.

[0108] This involves a depth parameter, which corresponds to each contour pixel and represents the actual distance from the robot to the surface of the obstacle corresponding to that pixel.

[0109] By acquiring visual images and performing masking operations, the features of corresponding pixels of obstacles can be enhanced in a targeted manner, while the interference of non-obstacles can be weakened. This makes the boundary between obstacles and the background in the mask image clearer, reducing the interference of non-obstacle pixels on subsequent recognition from the source, thereby improving the accuracy of obstacle detection.

[0110] As an optional embodiment, determining multiple candidate path regions for the robot based on obstacle space parameters and obstacle hardness parameters includes: determining the robot's action space, wherein the action space includes multiple spatial units, and the action space is the three-dimensional space corresponding to the robot's target action; determining obstacle features corresponding to the multiple spatial units based on the obstacle space parameters and obstacle hardness parameters, wherein the corresponding obstacle features represent the characteristics of the obstacles in the corresponding spatial units; and determining multiple candidate path regions for the robot based on the obstacle features corresponding to the multiple spatial units.

[0111] This embodiment describes the specific steps for determining multiple candidate path regions for the robot based on obstacle spatial parameters and obstacle hardness parameters.

[0112] This involves the motion space, which is the three-dimensional spatial range involved when the robot performs a target action. The motion space provides a basic three-dimensional spatial framework for the robot to plan its motion path and is the spatial boundary of path planning. For example, if the robot's target action is to move forward 10 meters, the motion space can be a three-dimensional spatial region centered on the robot's current position and extending forward 10 meters.

[0113] This involves multiple spatial units, which are small spatial units obtained by dividing the action space. Each spatial unit is a small three-dimensional region. For example, the action space can be divided into 10cm×10cm×10cm cube units, with each unit representing a small area within the action space.

[0114] This involves target actions, which are the specific tasks or motion goals that the robot needs to complete. Target actions determine the range and shape of the motion space. Target actions can be forward movement, turning, grasping, etc., and different target actions correspond to different motion spaces.

[0115] This involves obstacle features corresponding to multiple spatial units. These obstacle features are characteristic information related to obstacles within each spatial unit, which may include the spatial distribution characteristics of obstacles (such as whether there are obstacles, the number and density of feature points of obstacles), and the hardness characteristics of obstacles (such as whether they are rock, wood or plastic), in order to assess the feasibility and risk level of movement in each spatial unit and thus determine candidate path areas.

[0116] First, the robot's motion space is defined, clearly defining its three-dimensional motion boundaries when performing target actions. Dividing the motion space into multiple small spatial units breaks down a large environment into easily analyzable smallest units, avoiding ambiguity in obstacle feature judgment due to an excessively large spatial range. Then, based on obstacle spatial parameters (clarifying obstacle location and shape) and hardness parameters (clarifying obstacle hardness), the obstacle features of each unit are determined, accurately judging whether obstacles exist in each unit and the level of obstacle risk, thereby filtering out units with feasible movement and low risk. Finally, these units are merged to form multiple candidate path regions. The defined motion space ensures that all candidate paths are within the robot's motion capabilities, preventing paths that exceed its motion limits. Furthermore, the comprehensive integration of obstacle features eliminates high-risk areas, making path planning more closely aligned with the actual environment. At the same time, the setting of multiple candidate regions provides multiple feasible path selections, avoiding the limitations of a single path, ultimately achieving a dual improvement in the accuracy and safety of path planning.

[0117] As an optional embodiment, multiple candidate path regions of the robot are determined based on the obstacle features corresponding to the multiple spatial units, including: determining the adjacent units corresponding to the multiple spatial units; for any target spatial unit among the multiple spatial units, determining the similarity index between the target spatial unit and its adjacent units based on the obstacle features of the target spatial unit and the obstacle features of its adjacent units; determining the result of whether to perform a merging operation corresponding to the target spatial unit based on the similarity index, wherein the merging operation is used to merge the target spatial unit with its adjacent units; determining the result of whether to perform a merging operation corresponding to the target spatial unit, using the same method as determining the result of whether to perform a merging operation corresponding to the target spatial unit; and determining multiple candidate path regions of the robot based on the result of whether to perform a merging operation corresponding to the multiple spatial units.

[0118] This embodiment describes the specific steps for determining multiple candidate path regions of the robot based on the obstacle features corresponding to multiple spatial units.

[0119] This involves adjacent units, which are other spatial units that are adjacent to the corresponding spatial unit among multiple spatial units.

[0120] This involves the target spatial unit, which is the spatial unit among multiple spatial units that is currently being analyzed for merging with adjacent units. In the merging process, one spatial unit can be selected as the target first, its similarity index with adjacent units can be calculated, and a merging decision can be made. This method can then be replicated to all other spatial units to ensure that each unit completes the merging decision. For example, (X1,Y1,Z1) can be selected as the target spatial unit, and then spatial units such as (X1,Y1,Z2), (X2,Y1,Z1), etc., can be analyzed sequentially.

[0121] This involves a similarity index, which quantifies the degree of similarity in obstacle features between a target spatial unit and its neighboring units. This similarity index is used to determine whether a target spatial unit and its neighboring units have similar obstacle features, thereby deciding whether to merge them.

[0122] This involves a merging operation, which is the operation of merging the target spatial unit with adjacent units into a larger spatial unit (i.e., a spatial region).

[0123] This involves determining the outcome, which is based on a similarity index to decide whether to perform a merge operation.

[0124] By identifying adjacent units of a spatial unit, the comparison object for each unit can be clearly defined, thus defining the scope for subsequent similarity judgment. Calculating the similarity index between the target unit and its adjacent units can accurately quantify the similarity of their obstacle features (such as obstacle distribution and hardness), avoiding subjective judgment errors and effectively identifying units with similar obstacle features. The decision to merge units based on the similarity index allows units with consistent features to be integrated into a larger area, reducing the number of independent units that need to be analyzed during path planning and simplifying the planning process. Simultaneously, the unified obstacle features of the merged area prevent the robot from frequently adjusting its posture during the path, improving path coherence. Multiple merged areas constitute a diverse candidate path region, covering feasible areas within the robot's action space. Furthermore, the obstacle feature-based selection ensures path safety, ultimately enabling path planning to accurately adapt to the environment while providing the robot with multiple options, enhancing its adaptability to complex environments.

[0125] As an optional embodiment, before obtaining the obstacle hardness parameters of the robot, the method further includes: determining the force data of the robot within a target time period; determining the force change parameters corresponding to multiple sub-time periods based on the force data, wherein the target time period includes multiple sub-time periods; determining the force intensity parameters corresponding to the robot based on the force change parameters corresponding to the multiple sub-time periods; and determining the obstacle hardness parameters corresponding to the robot based on the force intensity parameters.

[0126] This embodiment describes the specific steps before obtaining the obstacle hardness parameters of the robot.

[0127] This involves a target time period, which is the complete time interval during which the robot comes into contact with the obstacle and is used to collect force data.

[0128] This includes force data, which is the data on the external forces exerted on the robot during the target time period, reflecting the physical interaction between the robot and the environment (such as obstacles).

[0129] This involves force variation parameters, which describe the changing patterns of force signals within each sub-time period. These parameters include the average force variation curve, the first-order difference curve, and the second-order difference curve. The average curve represents the average force signal from all sensor units within the sub-time period, reflecting the overall force level during that period. The first-order difference curve reflects the rate of force change, and the second-order difference curve reflects the acceleration of force change. Together, these three curves constitute the dynamic characteristics of force variation, providing a basis for subsequent calculations of force intensity parameters.

[0130] This involves multiple sub-time periods, which are multiple consecutive small time intervals divided by a preset window length for the target time period.

[0131] This involves a force intensity parameter, which is used to quantify the force intensity of the robot within a target time period. This force intensity parameter can be a trend feature vector that reflects the force trend.

[0132] The target time period is divided into multiple sub-time periods, and the force change parameters of each sub-time period are determined. Force analysis of the robot is realized from the time dimension, which can capture the force change characteristics of the robot in more detail. Furthermore, by integrating the force change parameters of multiple sub-time periods, the force intensity parameters can be accurately determined to more comprehensively reflect the interaction force between the robot and the obstacle, thereby achieving accurate determination of the obstacle hardness parameters.

[0133] Based on the above embodiments and optional embodiments, an optional implementation method is provided, which is described in detail below.

[0134] In related technologies, to ensure that robots can complete motion operations according to task requirements while maintaining posture stability during movement, avoiding collisions with obstacles in the environment, and guaranteeing task execution efficiency and robot safety, it is necessary to determine a reasonable motion control strategy. However, in related technologies, when controlling the motion of robots, there is a technical problem that a fixed motion control strategy is difficult to adapt to complex unstructured environments, leading to insufficient stability and reliability of robot motion.

[0135] There is currently no effective solution to the above problems.

[0136] In view of this, an optional embodiment of the present invention provides a method for determining the motion control strategy of a robot, which can effectively solve the above-mentioned technical problems.

[0137] Figure 2 This is a flowchart of a method for determining the motion control strategy of a robot according to an optional embodiment of the present invention. Figure 3 This is a schematic diagram of the motion control system structure in an optional embodiment of the present invention, such as... Figure 2 as well as Figure 3 As shown below, a detailed description will be provided.

[0138] S1. Obtain the robot's obstacle space parameters, obstacle hardness parameters, and joint motion parameters;

[0139] Before acquiring the robot's obstacle space parameters, the process also includes determining the robot's obstacle space parameters, that is, identifying the spatial location information of obstacles based on visual image data. This specifically includes the following steps:

[0140] A1. Determine the robot's visual image;

[0141] Furthermore, the visual image (i.e., visual image data) can be a color image.

[0142] A2. Perform a masking operation on multiple image pixels of the visual image to obtain a mask image corresponding to the visual image. The masking operation is used to perform feature enhancement processing on the first pixel among the multiple image pixels and feature reduction processing on the second pixel among the multiple image pixels. The first pixel is the pixel among the multiple image pixels that corresponds to the obstacle, and the second pixel is used to represent the pixel among the multiple image pixels that corresponds to the non-obstacle.

[0143] Taking a color image as an example, semantic segmentation is performed on the color image (same as the masking operation described above). Obstacle pixel regions are marked, for example, a binary mask image is obtained (same as the mask image described above), and the edge contour features of the obstacle regions are extracted to determine the contour pixel positions. Specifically, this can be achieved in the following way:

[0144] Semantic segmentation is performed using the DeepLab V3+ semantic segmentation network (with ResNet101 as the backbone and Pascal VOC dataset as the training set). After processing by the DeepLab V3+ network, each pixel in the color image can be divided into two categories: obstacles and non-obstacles. A binary mask image is output, in which the pixel values ​​of obstacle regions (same as the first pixel mentioned above) are marked as 1, and the rest (same as the second pixel mentioned above) are marked as 0.

[0145] A3. Based on the mask image, determine the contour parameters corresponding to the obstacle. The contour parameters include multiple contour pixels and position parameters corresponding to each contour pixel.

[0146] Taking a color image as an example, semantic segmentation is performed on the color image (same as the masking operation described above). Obstacle pixel regions are marked, for example, a binary mask image is obtained (same as the mask image described above), and the edge contour features of the obstacle regions are extracted (same as the contour parameters described above). The positions of the contour pixels are determined (same as the position parameters corresponding to the multiple contour pixels described above). Specifically, this can be achieved in the following way:

[0147] The Canny edge detection algorithm is used to extract the contour pixel positions of the obstacle region and obtain the two-dimensional boundary coordinates of the obstacle region (corresponding to the position parameters of the multiple contour pixels mentioned above).

[0148] For example, the two-dimensional contour points of the obstacle (similar to the multiple contour pixels mentioned above) may include specific pixel coordinates such as (450,320), (455,321), (460,322), ...

[0149] A4. Based on the visual image, determine the depth parameters corresponding to each of the multiple contour pixels;

[0150] Based on the depth image data matrix in the visual image data (same as the visual image described above), the precise distance data (same as the depth parameters described above) corresponding to the contour pixel positions is extracted. Specifically, at the determined contour pixel positions, contour feature points are densely sampled at fixed pixel intervals (e.g., every 5 pixels); for example, the feature points include (450, 320), (455, 321), (460, 322), etc.

[0151] For each contour feature point (i.e., each contour pixel) among multiple contour pixels, the precise distance data of the corresponding position is directly obtained through the depth image data matrix; for example, the depth distance corresponding to contour feature point (450,320) is 1.05m, (455,321) is 1.08m, (460,322) is 1.10m, etc., forming specific distance data.

[0152] A5. Based on the contour parameters and the depth parameters corresponding to multiple contour pixels, determine the obstacle space parameters corresponding to the robot.

[0153] Using the two-dimensional pixel coordinates of feature points (same as the above multiple contour pixels) (same as the above contour parameters, specifically, the position parameters corresponding to the multiple contour pixels included in the contour parameters) and precise distance data (same as the above depth parameters), the three-dimensional spatial coordinate position of the obstacle is calculated by combining the perspective projection model, and the spatial position information of the obstacle (same as the above obstacle spatial parameters) is generated.

[0154] The two-dimensional pixel coordinates (u, v) of the contour feature points are combined with the precise distance data D, based on intrinsic parameters calibrated by the visual sensor, such as focal length. =920 pixels =920 pixels, image principal point ( , (640, 360) is used to calculate the three-dimensional spatial coordinates of obstacle feature points using a perspective projection model.

[0155] Specifically, using the calibrated principal point coordinates of the image ( , As an example, the specific calculation method for three-dimensional spatial coordinates (X,Y,Z) is as follows:

[0156]

[0157]

[0158]

[0159] For example, with the feature point (450, 320) at a distance of 1.05m, the coordinates of the principal point of the calibrated image are ( , Taking (e.g.) as an example, the specific calculated three-dimensional coordinate position is as follows:

[0160]

[0161]

[0162]

[0163] Similarly, by calculating the three-dimensional coordinates of all contour feature points, the spatial location information of the obstacle can be clearly obtained.

[0164] Therefore, determining the robot's obstacle space parameters, that is, identifying the spatial location information of obstacles, includes:

[0165] Semantic segmentation is performed on the color images in the visual image data to mark the pixel regions of obstacles and extract the edge contour features of the obstacle regions to determine the position of the contour pixels. Based on the depth image data matrix in the visual image data, the precise distance data corresponding to the contour pixel positions is extracted. Using the two-dimensional pixel coordinates of the feature points and the precise distance data, combined with the perspective projection model, the three-dimensional spatial coordinate position of the obstacle is calculated to generate the spatial position information of the obstacle.

[0166] Before obtaining the obstacle hardness parameters of the robot, the process also includes determining the obstacle hardness parameters of the robot, specifically including:

[0167] B1. Determine the force data of the robot within the target time period;

[0168] Furthermore, the force data can be force signal data.

[0169] B2. Based on the force data, determine the force change parameters corresponding to multiple sub-time periods, wherein the target time period includes multiple sub-time periods;

[0170] Taking the force data as an example, the force signal data is divided into multiple time domain windows (similar to the multiple sub-time periods mentioned above) according to a preset window length.

[0171] Furthermore, this force variation parameter, including the average force variation curve, can be determined in the following way:

[0172] The average force signal of all sensor units within each window (as in each sub-time period above) is calculated point by point to obtain the average curve of force change.

[0173] B3. Based on the force change parameters corresponding to multiple sub-time periods, determine the force intensity parameters corresponding to the robot;

[0174] Furthermore, taking the force variation parameter as the force variation average curve as an example, the force variation average curve is processed by first-order difference and second-order difference to obtain first-order and second-order difference curves respectively, and the second-order difference curve is further divided into multiple trend sub-windows.

[0175] Perform linear trend fitting on the second-order difference data within the sub-window, calculate the slope value of the trend fitting for each sub-window, and sum the absolute values ​​of all slope values ​​one by one to construct a trend feature vector (same as the stress intensity parameter mentioned above).

[0176] B4. Based on the force strength parameters, determine the obstacle hardness parameters corresponding to the robot.

[0177] Furthermore, when the obstacle stiffness parameter includes the surface stiffness category of the obstacle, taking the force strength parameter as a trend feature vector as an example, the trend feature vector is dynamically time-warped and matched with the pre-built standard stiffness trend database, and combined with the confidence of historical stiffness category identification, the surface stiffness category of the obstacle is determined by weighting.

[0178] Visual image data and force signal data can be acquired in real time by the robot using its own color image and depth information (RGB-D) visual sensors and flexible capacitive tactile sensors.

[0179] For visual image data, it specifically includes two parts: color (RGB) images and depth images;

[0180] in:

[0181] A color image is a two-dimensional pixel matrix containing the surface texture and color of obstacles;

[0182] The depth image is a depth matrix that corresponds to each pixel of the color image. The value of each matrix element is the actual distance from the robot to the surface of the object represented by the corresponding pixel, in meters (m).

[0183] For force signal data: it is acquired in real time by a flexible capacitive tactile sensor array installed at the end of the robot's forelimb; the tactile sensor array contains multiple independent sensor units, each of which outputs a voltage value (in volts, V) that changes with pressure after being subjected to force, and each unit outputs independently and can reflect the real-time contact pressure changes.

[0184] Then, the spatial location information of the obstacle is identified based on the visual image data, and the surface stiffness category of the obstacle is determined based on the force signal data.

[0185] Furthermore, when the obstacle stiffness parameter includes the obstacle's surface stiffness category, and the force variation parameter includes the average force variation curve, the first-order difference curve, and the second-order difference curve, determining the obstacle's surface stiffness category includes:

[0186] The force signal data is divided into multiple time-domain windows (i.e., multiple sub-time periods) according to a preset window length. The force signals from all sensor units within each window (i.e., each sub-time period) are averaged point-by-point to obtain the average force change curve. Specifically, the flexible tactile sensor array on the robot's forelimb surface outputs the force signal data (i.e., force data) during the contact process in real time. The continuous contact process force signal data is divided into multiple clearly defined time-domain windows with a window length of 200ms, each containing 20 sampling points.

[0187] In practice, the voltage signal data output by all sensor units (which can be 16) in each window at each moment is averaged point by point to obtain the average curve of the force change in the window.

[0188] The average curve of force change is processed by first-order and second-order difference to obtain first-order and second-order difference curves respectively, and the second-order difference curve is further divided into multiple trend sub-windows.

[0189] It should be noted that: the first-order difference curve reflects the velocity characteristics of the force change, while the second-order difference curve reflects the acceleration characteristics of the force change.

[0190] Subsequently, the second-order difference curve is divided into multiple trend sub-windows with a fixed length (such as 4 consecutive points) for subsequent trend analysis;

[0191] Perform linear trend fitting on the second-order difference data within the sub-window, calculate the slope value of the trend fitting for each sub-window, and sum the absolute values ​​of all slope values ​​one by one to construct a trend feature vector.

[0192] Specifically, the least squares method is used to fit the linear trend to obtain the fitting slope value representing the force acceleration trend of the sub-window; then the absolute value of the slope value of each sub-window is taken and accumulated to obtain the trend feature vector (that is, the force intensity parameter) that can represent the overall force change trend intensity.

[0193] For example, if the slope values ​​of each sub-window in a certain window are 0.1, -0.05, 0.08, -0.12, and 0.06 respectively, then the value of the trend feature vector is the sum of the absolute values ​​of these slopes, which is 0.41.

[0194] The trend feature vector is dynamically time-warped and matched with a pre-built standard stiffness trend database. Combined with the confidence level of historical stiffness category identification, the surface stiffness category of the obstacle included in the obstacle hardness parameter is determined by weighting.

[0195] The aforementioned trend feature vectors are matched with a pre-built standard stiffness trend database using dynamic time warping (DTW) to determine the matching distance between the current trend feature vector and each standard trend. .

[0196] Combined with the historical recognition accuracy (i.e., historical confidence) of each obstacle stiffness category in the robot's historical task records To clearly calculate the final composite score for each category. The specific calculation formula is as follows:

[0197]

[0198] For example, if the trend feature vector matches the DTW distance of the rock category by 0.2, and the historical recognition accuracy of the rock category is 0.9, then the final comprehensive score for the rock category is calculated as follows:

[0199]

[0200] Select the stiffness category corresponding to the lowest overall score as the current defined surface stiffness category for the obstacle.

[0201] Therefore, when the obstacle stiffness parameter includes the obstacle's surface stiffness category, and the force variation parameter includes the average force variation curve, the first-order difference curve, and the second-order difference curve, determining the obstacle's surface stiffness category includes:

[0202] The force signal data is divided into multiple time-domain windows according to a preset window length, and the force signals of all sensor units in each window are averaged point by point to obtain the average force change curve. The average force change curve is processed by first-order and second-order difference to obtain first-order and second-order difference curves, respectively. The second-order difference curve is further divided into multiple trend sub-windows. Linear trend fitting is performed on the second-order difference data in the sub-window, and the slope value of the trend fitting for each sub-window is calculated. The absolute values ​​of all slope values ​​are accumulated one by one to construct a trend feature vector. The trend feature vector is dynamically time-warped and matched with a pre-constructed standard stiffness trend database. Combined with the confidence level of historical stiffness category identification, the surface stiffness category of the obstacle is determined by weighting.

[0203] S2. Based on the obstacle space parameters and obstacle hardness parameters, determine multiple candidate path zones for the robot;

[0204] Specifically, S2 also includes:

[0205] S21. Determine the robot's motion space, wherein the motion space includes multiple spatial units, and the motion space is the three-dimensional space corresponding to the robot's target motion;

[0206] Taking the target action as a forward motion as an example, a virtual three-dimensional space grid (same as the motion space mentioned above) is constructed based on the spatial range of the robot's forward direction.

[0207] Furthermore, constructing a virtual three-dimensional spatial grid based on the spatial range of the robot's forward direction is as follows: constructing a three-dimensional spatial grid based on the effective spatial range of the robot's forward movement.

[0208] For example, the three-dimensional space within a 3m radius in front of the robot can be discretized according to a set resolution (e.g., 10cm) to obtain a large number of regular grid cells. Each grid cell is identified by a specific three-dimensional spatial coordinate. For example, the grid cell coordinates (1,3,2) correspond to the spatial position of 1 cell in the positive X-axis direction, 3 cells in the positive Y-axis direction, and 2 cells in the positive Z-axis direction of the three-dimensional coordinate system.

[0209] S22. Based on the obstacle spatial parameters and obstacle hardness parameters, determine the obstacle features corresponding to multiple spatial units respectively, where the corresponding obstacle features represent the characteristics of the obstacles in the corresponding spatial unit.

[0210] Among them, the obstacle features corresponding to multiple spatial units (i.e., multiple grid units) include the spatial distribution features of obstacles within each grid unit and the surface stiffness category of obstacles.

[0211] Furthermore, taking the spatial position distribution features of obstacles within each grid cell (i.e., each spatial cell) corresponding to multiple spatial cells as an example, for multiple spatial grid cells (same as the above multiple spatial cells) included in the three-dimensional spatial grid, multiple contour pixels (i.e., all feature points of the three-dimensional spatial position of the obstacle) are mapped to the corresponding spatial grid cells one by one, and the spatial position distribution features of obstacles within each grid cell are statistically analyzed, including the number and density of spatial position feature points.

[0212] For example, all the calculated 3D spatial location feature points of obstacles (similar to the multiple contour pixels mentioned above) are mapped one by one to the corresponding spatial grid cells. The mapping method is as follows: a one-to-one correspondence is established between the 3D coordinates (X, Y, Z) of the feature points (i.e., contour pixels) and the coordinates of the grid cells, clarifying that each feature point (i.e., each contour pixel) belongs to a specific grid cell. Then, the number of feature points contained in each grid cell is counted. And calculate the feature point density of each grid cell. The specific calculation formula is as follows:

[0213]

[0214] in:

[0215] The volume of a single grid cell (i.e., a single spatial cell).

[0216] Grid cell volume It could be: .

[0217] For example, for a grid cell (1,3,2), if the number of feature points... If the value is 5, then the feature point density .

[0218] S23. Determine the adjacent units corresponding to the multiple spatial units respectively;

[0219] S24. For any target spatial unit among multiple spatial units, determine the similarity index between the target spatial unit and its adjacent units based on the obstacle characteristics of the target spatial unit and the obstacle characteristics of the adjacent units corresponding to the target spatial unit.

[0220] Furthermore, the similarity index includes a first index and a second index, wherein the first index is used to represent the difference in feature point (i.e., contour pixel) density between any spatial cell and its adjacent cells, and the category index is used to represent the degree of similarity of obstacle surface stiffness category between any spatial cell and its adjacent cells.

[0221] S25. Based on the similarity index, determine the result of whether to perform a merging operation corresponding to the target spatial unit, wherein the merging operation is used to merge the target spatial unit with the adjacent units of the target spatial unit;

[0222] Furthermore, taking a grid cell as an example, spatially adjacent grid cells (similar to the adjacent cells mentioned above) are merged. The merging condition (i.e., the condition for performing the merging operation) is as follows:

[0223] The first index is less than a preset difference threshold, that is, the difference in feature point density between adjacent grid cells (i.e., any spatial cell and the adjacent cells corresponding to that spatial cell) is less than a preset difference threshold (e.g., 1000 cells / m³).

[0224] Furthermore, the second index is greater than or equal to a preset similarity threshold, for example, the obstacle surface stiffness categories are consistent.

[0225] S26. Using the method of determining whether to perform a merging operation corresponding to the target spatial unit, determine whether to perform a merging operation corresponding to other spatial units among multiple spatial units other than the target spatial unit.

[0226] S27. Based on the determination results of whether to perform the merging operation corresponding to multiple spatial units, determine multiple candidate path regions for the robot.

[0227] Through a gradual recursive merging process, several path candidate regions with clear regional boundaries and consistent internal characteristics are eventually formed (i.e., multiple candidate path regions).

[0228] For example, grid cell (1,3,2) has a density of 5000 cells / m³ and a stiffness category of "rock". Its density difference with the neighboring grid cell (i.e., adjacent cell) (1,3,3) is 500 cells / m³ (below the threshold of 1000 cells / m³), and the stiffness categories are the same. Therefore, the two cells (i.e., this grid cell and its adjacent cells) are merged into the same path candidate area (i.e., the same candidate path area).

[0229] Therefore, based on the obstacle spatial parameters and obstacle hardness parameters, multiple candidate path regions for the robot are determined, that is, multiple path candidate regions are divided as follows:

[0230] A virtual three-dimensional spatial grid is constructed within the spatial range of the robot's forward direction; all feature points of the three-dimensional spatial position of the obstacle are mapped to the corresponding spatial grid cell one by one, and the spatial position distribution characteristics of the obstacle in each grid cell are statistically analyzed, including the number and density of spatial position feature points; based on the spatial position distribution characteristics of the obstacle in each grid cell and the surface stiffness category of the obstacle, spatially adjacent and similar grid cells are merged to form multiple path candidate areas.

[0231] S3. Determine the motion risk parameters corresponding to the multiple candidate path areas, where the corresponding motion risk parameters represent the degree of risk of the robot moving in the corresponding candidate path area;

[0232] Specifically, S3 also includes:

[0233] S31. Determine the robot's historical collision data;

[0234] S32. Based on historical collision data, determine the collision damage index corresponding to multiple obstacle hardness categories, where the obstacle hardness parameter includes multiple obstacle hardness categories.

[0235] The collision damage index for each obstacle hardness category (i.e., each stiffness category) can be the collision damage coefficient.

[0236] Furthermore, collision damage records (i.e., historical collision data) of obstacles of different stiffness categories are extracted from the robot's historical collision data, and the collision damage coefficient of each stiffness category is calculated. This can be determined in the following way:

[0237] Based on historical collision data of obstacles of different stiffness categories, including peak force data of each robot joint, the collision damage coefficient for each stiffness category is calculated. The formula is:

[0238]

[0239] in:

[0240] The peak force of the j-th collision experienced by the robot's joints during the historical collisions;

[0241] This represents the total number of historical collision events.

[0242] S33. For any target candidate path region among multiple candidate path regions, determine the obstacle distribution characteristics corresponding to the multiple obstacle hardness categories respectively;

[0243] Among them, the characteristics of obstacle distribution include the spatial non-uniformity index.

[0244] Spatial statistical analysis was performed on the spatial distribution of obstacles within each path candidate area (i.e., each candidate path region) to calculate the spatial distribution non-uniformity index. :

[0245]

[0246] in:

[0247] For the first The Euclidean distance from each feature point (i.e., contour pixel) to the center of the region;

[0248] This is the average distance of all feature points;

[0249] This represents the total number of feature points (i.e., the total number of contour pixels).

[0250] S34. Based on the collision damage index and obstacle distribution characteristics corresponding to multiple obstacle hardness categories, determine the motion risk parameters corresponding to the target candidate path area;

[0251] Taking the collision damage index as an example, the collision damage coefficient can be achieved in the following way:

[0252] Based on the collision damage coefficient, a nonlinear risk weight function is constructed, and the corresponding stiffness risk weight is calculated using the following formula:

[0253]

[0254] in:

[0255] e is a natural constant;

[0256] These are nonlinear mapping coefficients, which depend on the structural stiffness characteristics of the robot;

[0257] Using a covariance fusion algorithm, the spatial inhomogeneity index and stiffness risk weight are fused to form the multimodal path risk features (i.e., motion risk parameters) for each candidate path region:

[0258]

[0259] in:

[0260] For sports risk parameters;

[0261] The variance of the spatial inhomogeneity index;

[0262] This represents the variance of the stiffness risk weight.

[0263] S35. By determining the motion risk parameters corresponding to the target candidate path area, the motion risk parameters corresponding to other candidate path areas besides the target candidate path area are determined.

[0264] Therefore, determining the motion risk parameters corresponding to multiple candidate path regions, that is, calculating the multimodal path risk characteristics of the path candidate regions, includes:

[0265] Spatial statistical analysis is performed on the spatial distribution of obstacles within the path candidate area to calculate the spatial distribution non-uniformity index; collision damage records of obstacles of different stiffness categories are extracted from the robot's historical collision data to calculate the collision damage coefficient of each stiffness category; based on the collision damage coefficient, a nonlinear risk weight function is constructed to calculate the corresponding stiffness risk weight; using the covariance fusion algorithm, the spatial non-uniformity index and stiffness risk weight are fused to form multimodal path risk features (i.e., motion risk parameters).

[0266] S4. Based on the joint motion parameters and obstacle stiffness parameters, determine the stability constraint parameters corresponding to the multiple candidate path regions respectively. The corresponding stability constraint parameters are used to represent the degree of constraint of the corresponding candidate path region on the stability of the robot's motion posture.

[0267] Furthermore, when the joint motion parameters include the robot's historical joint motion data, the final attitude stability constraint parameters corresponding to each path candidate region (i.e., multiple candidate path regions) are calculated based on the surface stiffness category of the path candidate region and the robot's historical joint motion data.

[0268] Specifically, S4 also includes:

[0269] S41. Based on the joint motion parameters, determine the joint angle parameters, joint load parameters, and joint state parameters corresponding to the robot.

[0270] This involves determining the joint angle parameters corresponding to the robot based on joint motion parameters, that is, extracting joint angle sequences from the robot's historical motion data. Angular velocity sequence and angular acceleration sequence The joint scheduling parameters include joint angle sequence, angular velocity sequence, and angular acceleration sequence.

[0271] For example, using the joint motion data recorded in the robot's historical motion database as a basis, specific joint angle sequences can be extracted. Angular velocity sequence and angular acceleration sequence .

[0272] S42. Based on the joint angle parameters, determine the motion posture characteristic parameters corresponding to the robot;

[0273] When the motion posture feature parameters include the posture dynamics feature matrix, based on the extracted joint angle sequence Angular velocity sequence and angular acceleration sequence The three sequences were aligned in the time domain, and the attitude dynamics feature matrix was constructed.

[0274] To ensure that the three sequences can be correlated at the same time point, they are first temporally aligned with the same sampling interval (e.g., 10 ms) to construct a unified attitude dynamics feature matrix. Its structure is clearly defined as follows:

[0275]

[0276] in:

[0277] For historical sampling moments, This represents the total number of samples.

[0278] They are time points The data includes the angle, angular velocity, and angular acceleration of the robot's joints.

[0279] For example, select the data from the first 3 sampling points of joint 1 in the historical motion:

[0280] Time 10ms: ;

[0281] Time 20ms: ;

[0282] Time 30ms: .

[0283] Then the attitude dynamics characteristic matrix for:

[0284]

[0285] S43. Based on the obstacle hardness parameters, determine the obstacle hardness category corresponding to each of the multiple candidate path regions;

[0286] S44. Based on the motion posture feature parameters and the obstacle hardness categories corresponding to the multiple candidate path regions, determine the initial constraint parameters corresponding to the multiple candidate path regions respectively.

[0287] Specifically, this is achieved through the following methods:

[0288] The attitude dynamics feature matrix (same as the motion attitude feature parameters mentioned above) and the obstacle surface stiffness category of the corresponding path candidate region (that is, the obstacle hardness category corresponding to each candidate path region) are input into the pre-constructed attitude stability inference network, and the initial attitude stability constraint parameters are output.

[0289] Furthermore, an attitude stability inference network based on a Long Short-Term Memory (LSTM) network is employed. Specifically, this attitude stability inference network uses the attitude dynamics feature matrix... The initial attitude stability constraint parameters are output by taking the obstacle surface stiffness category encoding vector (e.g., rock is coded as [1,0,0], wood as [0,1,0], and plastic as [0,0,1]) as input. (That is, the initial constraint parameters).

[0290] For example, for the input feature matrix The example values ​​of the initial stability constraint parameters output by the attitude stability inference network are: [1,0,0], corresponding to the category code [1,0,0] of the obstacle stiffness category "rock".

[0291]

[0292] S45. Based on the joint load parameters and joint state parameters, the initial constraint parameters corresponding to the multiple candidate path regions are modified respectively to obtain the stability constraint parameters corresponding to the multiple candidate path regions respectively.

[0293] Specifically, this is achieved through the following methods:

[0294] Acquire real-time data for each joint of the robot, including load data (i.e., joint load parameters) and motion state data (i.e., joint state parameters), and calculate the error between the real-time data and the initial posture stability constraint parameters (the same as the initial constraint parameters mentioned above).

[0295] The load data includes load torque data. Furthermore, when the motion state data includes real-time joint angles, angular velocities, and angular accelerations, the real-time data is converted into real-time parameters. The real-time data conversion method is consistent with the initial parameters, and then the error vector between the real-time parameters and the initial attitude stability constraint parameters is calculated. The formula is:

[0296]

[0297] For example, if Real-time parameters of robot joints If the range is [0.80, 0.70, 0.60], then the calculation error value is:

[0298]

[0299] Using the error value, the initial attitude stability constraint parameters are dynamically iteratively corrected using the least squares method to obtain the final attitude stability constraint parameters (i.e., stability constraint parameters).

[0300] Furthermore, based on the joint load parameters and joint state parameters, the initial constraint parameters corresponding to each of the multiple candidate path regions are modified to obtain the stability constraint parameters corresponding to each of the multiple candidate path regions. This can also be achieved by using the error values ​​calculated in real time. The input is fed into the dynamic parameter correction model, and the initial parameters (i.e., the initial constraint parameters) are iteratively corrected using the least squares method. The correction formula is as follows:

[0301]

[0302] in:

[0303] for Attitude stability constraint parameters at time t;

[0304] The Jacobian matrix is ​​constructed based on historical data and is obtained through offline calculation of historical motion data. It represents the sensitivity of attitude parameter changes to error changes.

[0305] For example, at the initial moment A clear Jacobian matrix for:

[0306]

[0307] Substitute the predetermined error value ],calculate The corrected final attitude stability constraint parameters (i.e., stability constraint parameters) are obtained as follows:

[0308]

[0309] Therefore, calculating the attitude stability constraint parameters (i.e., the stability constraint parameters) includes:

[0310] Joint angle sequences, angular velocity sequences, and angular acceleration sequences are extracted from the robot's historical motion data. These three sequences are then time-domain aligned to construct an attitude dynamics feature matrix. The attitude dynamics feature matrix, along with the surface stiffness categories of obstacles in the corresponding path candidate regions, is input into a pre-constructed attitude stability inference network, which outputs initial attitude stability constraint parameters. Load data and motion state data of each joint of the robot are acquired in real time, and the error value between the real-time data and the initial attitude stability constraint parameters is calculated. Using the error value, the initial attitude stability constraint parameters are dynamically iteratively corrected using the least squares method to obtain the final attitude stability constraint parameters.

[0311] S5. Based on the motion risk parameters and stability constraint parameters corresponding to the multiple candidate path regions, determine the target path region from the multiple candidate path regions;

[0312] S6. Determine the robot's motion control strategy based on the motion risk parameters of the target path area.

[0313] Furthermore, the motion risk parameters and stability constraint parameters corresponding to multiple candidate path regions are input into a pre-built deep reinforcement learning model (hereinafter referred to as the deep reinforcement learning model) to generate the robot's initial control strategy. The initial control strategy is then corrected based on the real-time updated path risk features. That is, multimodal path risk features and attitude stability constraint parameters are input into the pre-built deep reinforcement learning model to generate an initial motion strategy sequence, and the motion strategy sequence is corrected based on the real-time updated path risk features. The specific steps are as follows:

[0314] Furthermore, it also includes training a deep reinforcement learning model, using the trained deep learning reinforcement model to determine the target path region from multiple candidate path regions based on the motion risk parameters and stability constraint parameters corresponding to the multiple candidate path regions, and determining the robot's motion control strategy based on the motion risk parameters of the target path region.

[0315] From the robot's historical motion data, we extract the historical multimodal path risk feature sequence, the historical posture stability constraint parameter sequence, and the corresponding historical robot joint motion command sequence, respectively.

[0316] That is, extracting data from the robot's historical motion database, including:

[0317] Historical moment path risk feature vector For example, [0.32, 0.45, 0.28];

[0318] Attitude stability constraint parameter vector;

[0319] Attitude stability constraint parameter vector For example, [0.80, 0.70, 0.60];

[0320] And the robot's historical joint movement command sequence For example, [Joint 1 angle command, Joint 2 angle command, ...;

[0321] The state space of the model is constructed using the multimodal path risk feature sequence and the posture stability constraint parameter sequence, and the action space of the model is constructed using the robot joint motion command sequence, thus generating a state-action training dataset.

[0322] The data from the two sequences above are combined to form a state vector. ;

[0323] The robot's historical joint motion command sequence is a motion vector. For example, a certain historical state vector:

[0324]

[0325] The corresponding action vector is:

[0326]

[0327] This is how the state-action pair training dataset is constructed;

[0328] Furthermore, deep reinforcement learning models can also employ the Deep Deterministic Policy Gradient (DDPG) reinforcement learning algorithm, where the model structure of a deep reinforcement learning model includes an Actor network and a Critic network.

[0329] The Actor network specifically adopts a three-layer fully connected neural network architecture, including:

[0330] The input layer has 6 dimensions, corresponding to path risk features (spatial inhomogeneity index, stiffness risk weight, fusion risk value, a total of 3 dimensions) and attitude stability constraint parameters (angle constraint, angular velocity constraint, angular acceleration constraint, a total of 3 dimensions).

[0331] The hidden layer consists of two layers, each containing 256 neurons, and the ReLU (Rectified Linear Unit) activation function is used between the hidden layers;

[0332] The output layer uses the tanh activation function and has an output dimension of 3, which corresponds to the continuous motion angle commands of robot joints 1-3. The specific command range is mapped to the range [-1, 1] by the tanh function and further linearly mapped to the actual joint command angle (e.g., [-0.5 rad, 0.5 rad]).

[0333] The Critic network specifically employs a two-stream network architecture with joint state-action input, including:

[0334] The state input stream receives a state vector with a dimension of 6, which is then passed through a fully connected layer with 256 neurons and the activation function is ReLU.

[0335] The action input stream receives an action vector of dimension 3, which is then passed through a fully connected layer with 128 neurons and the activation function is ReLU.

[0336] Specifically, the outputs of the two input streams, state and action, are concatenated and then passed through a fully connected layer with 256 neurons and the activation function is ReLU.

[0337] The final output is a scalar Q-value, used to evaluate the value of the current state-action pair;

[0338] Furthermore, the action policy value function of the deep reinforcement learning model is expressed as follows:

[0339]

[0340] in:

[0341] Value of action strategy;

[0342] This is the current state vector, which includes path risk features and attitude stability constraint parameters;

[0343] This is a motion vector, representing the robot's joint movement commands;

[0344] These are the parameters of the action strategy value network;

[0345] For instant rewards;

[0346] This is a discount factor (e.g., 0.99).

[0347] and For example, weighting coefficients ;

[0348] This is the next state vector;

[0349] This is the next action vector.

[0350] for It can be determined in the following ways:

[0351]

[0352] in:

[0353] For path risk;

[0354] It is the sum of the absolute values ​​of the errors between the current real-time state parameters and the attitude stability constraint parameters.

[0355] Furthermore, an experience replay mechanism is used to randomly sample state-action training samples, and the stochastic gradient descent algorithm is used to iteratively optimize the action policy value network parameters. The experience replay mechanism specifically includes:

[0356] Construct an experience pool of size 50,000, randomly select 128 state-action sample pairs each time, and perform explicit stochastic gradient descent training. The specific formula for network parameter optimization is as follows:

[0357]

[0358] in:

[0359] The learning rate, such as 0.001;

[0360] For parameters The gradient.

[0361] Repeat the stochastic gradient descent training process described above until the action policy value function converges, resulting in a trained deep reinforcement learning model.

[0362] For example, once the change in the loss function falls below a preset threshold (e.g., 0.001), the action policy value function is considered to have converged, and the parameters of the trained deep reinforcement learning model are output. ;

[0363] Furthermore, after determining the robot's motion control strategy based on the motion risk parameters of the target path region, the process also includes revising the motion strategy sequence (i.e., the motion control strategy), as follows:

[0364] Based on real-time collected path risk characteristic data, calculate the real-time gradient of the path risk characteristics at the current moment compared to the path risk characteristics at the previous moment. The specific formula is as follows:

[0365]

[0366] in:

[0367] This represents the value of the path risk characteristic at the current moment.

[0368] The value of the path risk characteristic at the previous moment;

[0369] This is the timestamp for the current moment and the previous moment.

[0370] The instant reward is dynamically adjusted based on the real-time changes in the path risk characteristics:

[0371]

[0372] in:

[0373] The adjusted instant reward;

[0374] The original instant reward;

[0375] This is for adjusting the coefficient.

[0376] The value function of the action strategy is updated in real time using dynamically adjusted instantaneous rewards, expressed as:

[0377]

[0378] in:

[0379] This is the value of the updated action strategy.

[0380] The above process is performed in real time, that is, the dynamically corrected motion strategy sequence is regenerated in real time.

[0381] Therefore, the training process of a deep reinforcement learning model further includes:

[0382] From the robot's historical motion data, we extract the multimodal path risk feature sequence, the posture stability constraint parameter sequence, and the corresponding robot joint motion command sequence, respectively.

[0383] The state space of the model is constructed using the multimodal path risk feature sequence and the posture stability constraint parameter sequence, and the action space of the model is constructed using the robot joint motion command sequence, thus generating a state-action training dataset.

[0384] Define the action policy value function of a deep reinforcement learning model as follows:

[0385]

[0386] in: Value of action strategy; This is the current state vector, which includes path risk features and attitude stability constraint parameters; This is a motion vector, representing the robot's joint movement commands; These are the parameters of the action strategy value network; For instant rewards; This is a discount factor (e.g., 0.99). and For example, weighting coefficients ; This is the next state vector; For the next action vector;

[0387] An experience replay mechanism is used to randomly sample state-action training samples, and the stochastic gradient descent algorithm is used to iteratively optimize the action policy value network parameters. ;

[0388] Repeat the stochastic gradient descent training process described above until the action policy value function converges, resulting in a trained deep reinforcement learning model.

[0389] For example, suppose the next joint movement command in the original motion policy sequence is [0.18 rad, 0.14 rad, 0.09 rad]; after updating the motion policy value function, based on the new state input in real time (path risk 0.25 and posture constraint parameter 0.78), the model re-generates the corrected motion policy sequence. For example, the updated joint movement command is adjusted to [0.16 rad, 0.12 rad, 0.08 rad] to further reduce motion risk and improve posture stability.

[0390] Furthermore, after regenerating the dynamically corrected motion strategy sequence in real time, the process also includes temporal smoothing and redundant motion removal of the dynamically corrected motion strategy sequence based on robot joint continuity constraints to obtain an optimized motion control sequence, as detailed below:

[0391] C1. Extract the angle change data of each joint from the historical joint motion data;

[0392] Specifically, the continuous motion angle change data of each joint is extracted from the robot's historical motion database, that is, the angle difference of each joint between every two adjacent actions, in the following form:

[0393]

[0394] For example, the continuous angle change sequence of joint 1 in a certain historical motion data is [0.03 rad, 0.04 rad, 0.02 rad, 0.05 rad];

[0395] C2. Perform statistical analysis on historical angle change data to determine the allowable range of continuous joint movement angle changes;

[0396] Specifically, determining the permissible range of angle changes for continuous joint movement includes:

[0397] C21. Collect historical joint angle change data and create a data histogram;

[0398] Specifically, taking the historical angle change data of joint 1 [0.03, 0.04, 0.02, 0.05, 0.03, 0.04] as an example, the data is grouped with a step size of 0.01 rad, and a histogram is built to reflect the frequency of data distribution.

[0399] C22. Based on the data histogram, the probability density function is established using the kernel density estimation method, with the following formula:

[0400]

[0401] in:

[0402] This is an estimate of the target sample point x;

[0403] This is the i-th data sample;

[0404] n is the number of samples;

[0405] h is the kernel bandwidth ( );

[0406] This is the Gaussian kernel function.

[0407] C23. Determine the upper and lower boundaries of the confidence interval for the joint angle change value based on the probability density function:

[0408]

[0409] in:

[0410] It is a random variable;

[0411] , These are the upper and lower limits of the confidence interval;

[0412] The significance level (e.g., 0.05) is used to determine the probability density function. The probability threshold of the confidence interval, representing the random variable The value of falls within the interval [ , The maximum allowed probability value outside of [ ].

[0413] Furthermore, the upper and lower boundaries of the confidence interval are respectively defined as the lower and upper limits of the allowable range of angle changes during continuous joint movement.

[0414] Therefore, determining the permissible range of angle changes for continuous joint movement includes:

[0415] Historical joint angle variation data were collected and a data histogram was constructed. Based on the data histogram, a probability density function was established using the kernel density estimation method. The upper and lower boundaries of the confidence interval for the joint angle variation value were determined according to the probability density function. The upper and lower boundaries of the confidence interval were respectively determined as the lower and upper limits of the allowable range of angle variation for continuous joint movement.

[0416] C3. Calculate the real-time joint angle changes between adjacent movements in the dynamically corrected motion strategy sequence;

[0417] Specifically, for any adjacent actions in the dynamically corrected motion strategy sequence and Calculate the real-time joint angle change value:

[0418]

[0419] in:

[0420] This represents the real-time changes in the robot's joint angles;

[0421] For the dynamically corrected motion strategy sequence Action instructions at any given moment;

[0422] For the dynamically corrected motion strategy sequence Action instructions at any given moment.

[0423] For example: if the action at a certain moment is [0.20 rad, 0.15 rad, 0.10 rad], and the next moment is [0.22 rad, 0.17 rad, 0.12 rad], then the real-time angle change of joint 1 is 0.02 rad; if the real-time joint angle change value is less than the lower limit of the allowable interval, then the corresponding action is determined to be a redundant action and it is deleted from the motion strategy sequence.

[0424] For example, if the allowed continuous variation range of joint 1 is [0.02 rad, 0.05 rad], but the actual variation is only 0.015 rad, then the action is determined to be a redundant action and is explicitly deleted from the sequence to improve the simplicity and execution efficiency of the sequence.

[0425] C4. Execute the sequence after removing redundant actions. The filtering process is as follows:

[0426]

[0427] in:

[0428] This is the j-th filtered data (i.e., the joint angle command in the motion strategy).

[0429] This is the (j+1)th original data;

[0430] Let i be the i-th filter coefficient;

[0431] The value is half the width of the filter window (e.g., a value of 2 means that the window is two data points before and after the current point).

[0432] C5. Output the final optimized motion control sequence;

[0433] Taking the joint 1 sequence [0.20, 0.22, 0.21, 0.23, 0.22 rad] as an example, the smoothed sequence obtained through the above filtering process is [0.210, 0.216, 0.218, 0.218, 0.220 rad].

[0434] Therefore, the temporal smoothing and redundant action removal processes for the dynamically corrected motion strategy sequence include:

[0435] Extract angle change data for each joint from historical joint motion data; perform statistical analysis on the historical angle change data to determine the allowable range of continuous joint motion angle changes; calculate the real-time joint angle change values ​​between adjacent movements in the dynamically corrected motion strategy sequence; if the real-time joint angle change value is less than the lower limit of the allowable range, the corresponding movement is determined to be a redundant movement and removed from the motion strategy sequence; execute the sequence after removing redundant movements. The filter process outputs the final optimized motion control sequence.

[0436] The optimized motion control sequence output eliminates redundant movements with small angle changes, and after filtering, the motion sequence of each joint changes more continuously and smoothly, thereby improving the robot's execution efficiency and significantly reducing the risk of mechanical wear.

[0437] Furthermore, the above steps can be implemented using a biomimetic robot motion control system based on deep reinforcement learning, which includes:

[0438] The data acquisition module is used to acquire visual image data and force signal data; it identifies the spatial location information of obstacles based on visual image data and determines the surface stiffness category of obstacles based on force signal data.

[0439] The path segmentation module is used to divide multiple path candidate regions based on the spatial location information and surface stiffness category of obstacles, and to calculate the multimodal path risk characteristics of the path candidate regions.

[0440] The parameter calculation module is used to calculate the final attitude stability constraint parameters corresponding to each path candidate region based on the surface stiffness category of the path candidate region and the robot's historical joint motion data.

[0441] The model recognition module is used to input multimodal path risk features and attitude stability constraint parameters into a pre-built deep reinforcement learning model, generate an initial motion policy sequence, and correct the motion policy sequence based on the real-time updated path risk features.

[0442] The motion correction module is used to perform temporal smoothing and redundant motion removal on the dynamically corrected motion strategy sequence based on the robot joint continuity constraints, so as to obtain an optimized motion control sequence.

[0443] The above optional implementation methods can achieve at least the following beneficial effects:

[0444] (1) Compared with related technologies, this invention can simultaneously and accurately identify the spatial location information and surface stiffness category of obstacles by integrating visual image data and force signal data, which significantly improves the robot's comprehensive perception ability of unstructured environment and lays a solid foundation for subsequent efficient and safe path planning.

[0445] (2) Compared with related technologies, this invention effectively realizes the comprehensive real-time evaluation of path risk and robot posture state by constructing multimodal path risk features and combining them with real-time updated posture stability constraint parameters. This enables the robot to quickly and accurately adjust its motion strategy, significantly reducing motion risk and improving posture stability, thus ensuring the robot's motion safety in complex environments.

[0446] (3) Compared with related technologies, the present invention effectively improves the coherence and stability of the robot motion control strategy by dynamically correcting the motion strategy sequence and using temporal smoothing and redundant action screening, significantly reduces robot mechanical wear and motion execution costs, and achieves a comprehensive improvement in robot motion performance and task execution efficiency in complex unstructured environments.

[0447] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0448] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0449] Example 2

[0450] According to embodiments of the present invention, an apparatus for implementing the above-described method for determining the motion control strategy of a robot is also provided. Figure 4 This is a structural block diagram of a robot motion control strategy determination device according to an embodiment of the present invention, such as... Figure 4 As shown, the device includes: an acquisition module 402, a first determination module 404, a second determination module 406, a third determination module 408, a fourth determination module 410, and a fifth determination module 412. The device will be described in detail below.

[0451] The acquisition module 402 is used to acquire the robot's obstacle space parameters, obstacle hardness parameters, and joint motion parameters;

[0452] The first determining module 404 is connected to the aforementioned obtaining module 402 and is used to determine multiple candidate path regions for the robot based on obstacle space parameters and obstacle hardness parameters.

[0453] The second determining module 406 is connected to the first determining module 404 and is used to determine motion risk parameters corresponding to multiple candidate path areas respectively, wherein the corresponding motion risk parameters represent the degree of risk of the robot moving in the corresponding candidate path area.

[0454] The third determining module 408, connected to the second determining module 406, is used to determine the stability constraint parameters corresponding to the multiple candidate path regions based on the joint motion parameters and obstacle stiffness parameters. The corresponding stability constraint parameters are used to represent the degree of constraint of the corresponding candidate path region on the stability of the robot's motion posture.

[0455] The fourth determining module 410, connected to the third determining module 408, is used to determine the target path region from the multiple candidate path regions based on the motion risk parameters and stability constraint parameters corresponding to the multiple candidate path regions respectively.

[0456] The fifth determining module 412, connected to the fourth determining module 410, is used to determine the robot's motion control strategy based on the motion risk parameters of the target path area.

[0457] It should be noted that the above-mentioned acquisition module 402, first determination module 404, second determination module 406, third determination module 408, fourth determination module 410 and fifth determination module 412 correspond to steps S102 to S110 in the method for determining the motion control strategy of the robot. The multiple modules and the corresponding steps are the same in terms of the instances and application scenarios implemented, but are not limited to the content disclosed in the above embodiment 1.

[0458] Example 3

[0459] According to another aspect of the present invention, an electronic device is also provided, comprising: a processor; and a memory for storing processor-executable instructions, wherein the processor is configured to execute instructions to implement the robot motion control strategy determination method of any of the above embodiments.

[0460] Example 4

[0461] According to another aspect of the present invention, a computer-readable storage medium is also provided, which, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform the robot motion control strategy determination method described above.

[0462] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0463] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0464] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0465] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0466] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0467] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0468] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for determining the motion control strategy of a robot, characterized in that, include: Obtain the robot's obstacle space parameters, obstacle stiffness parameters, and joint motion parameters; Based on the obstacle spatial parameters and the obstacle hardness parameters, multiple candidate path regions for the robot are determined; Determine motion risk parameters corresponding to the plurality of candidate path regions, wherein the corresponding motion risk parameters represent the degree of risk of the robot moving in the corresponding candidate path region; Based on the joint motion parameters and the obstacle stiffness parameters, stability constraint parameters corresponding to the multiple candidate path regions are determined respectively, wherein the corresponding stability constraint parameters are used to represent the degree of constraint of the corresponding candidate path region on the motion posture stability of the robot. Based on the motion risk parameters and stability constraint parameters corresponding to the multiple candidate path regions, the target path region is determined from the multiple candidate path regions; Based on the motion risk parameters of the target path region, the motion control strategy of the robot is determined.

2. The method according to claim 1, characterized in that, Based on the joint motion parameters and the obstacle stiffness parameters, stability constraint parameters corresponding to the multiple candidate path regions are determined, including: Based on the joint motion parameters, determine the joint angle parameters, joint load parameters, and joint state parameters corresponding to the robot; Based on the joint angle parameters, determine the motion posture characteristic parameters corresponding to the robot; Based on the obstacle hardness parameters, determine the obstacle hardness category corresponding to each of the multiple candidate path regions; Based on the motion posture feature parameters and the obstacle hardness categories corresponding to the multiple candidate path regions, the initial constraint parameters corresponding to the multiple candidate path regions are determined respectively. Based on the joint load parameters and the joint state parameters, the initial constraint parameters corresponding to the plurality of candidate path regions are modified respectively to obtain the stability constraint parameters corresponding to the plurality of candidate path regions respectively.

3. The method according to claim 1, characterized in that, The determination of motion risk parameters corresponding to the plurality of candidate path regions includes: Determine the robot's historical collision data; Based on the historical collision data, collision damage indices corresponding to multiple obstacle hardness categories are determined, wherein the obstacle hardness parameters include multiple obstacle hardness categories. For any target candidate path region among the plurality of candidate path regions, determine the obstacle distribution characteristics corresponding to the plurality of obstacle hardness categories respectively; Based on the collision damage index and obstacle distribution characteristics corresponding to the multiple obstacle hardness categories, the motion risk parameters corresponding to the target candidate path area are determined; By determining the motion risk parameters corresponding to the target candidate path region, motion risk parameters are determined for each of the other candidate path regions besides the target candidate path region.

4. The method according to claim 1, characterized in that, Before obtaining the robot's obstacle space parameters, the following steps are also included: Determine the visual image of the robot; A masking operation is performed on multiple image pixels of the visual image to obtain a mask image corresponding to the visual image. The masking operation is used to perform feature enhancement processing on a first pixel among the multiple image pixels and feature reduction processing on a second pixel among the multiple image pixels. The first pixel is the pixel among the multiple image pixels that corresponds to an obstacle, and the second pixel is the pixel among the multiple image pixels that corresponds to a non-obstacle. Based on the mask image, the contour parameters corresponding to the obstacle are determined, wherein the contour parameters include multiple contour pixels and position parameters corresponding to the multiple contour pixels respectively; Based on the visual image, determine the depth parameters corresponding to the plurality of contour pixels respectively; Based on the contour parameters and the depth parameters corresponding to the plurality of contour pixels, the obstacle space parameters corresponding to the robot are determined.

5. The method according to claim 1, characterized in that, The step of determining multiple candidate path regions for the robot based on the obstacle spatial parameters and the obstacle hardness parameters includes: The robot's motion space is determined, wherein the motion space includes multiple spatial units, and the motion space is the three-dimensional space corresponding to the robot's target motion; Based on the obstacle spatial parameters and the obstacle hardness parameters, obstacle features corresponding to the plurality of spatial units are determined, wherein the corresponding obstacle features represent the characteristics of the obstacles in the corresponding spatial unit; Based on the obstacle features corresponding to the multiple spatial units, multiple candidate path regions of the robot are determined.

6. The method according to claim 5, characterized in that, The step of determining multiple candidate path regions for the robot based on the obstacle features corresponding to the multiple spatial units includes: Determine the adjacent units corresponding to the plurality of spatial units respectively; For any target spatial unit among the plurality of spatial units, a similarity index between the target spatial unit and its adjacent units is determined based on the obstacle features of the target spatial unit and the obstacle features of the adjacent units corresponding to the target spatial unit. Based on the similarity index, a determination result is made as to whether to perform a merging operation corresponding to the target spatial unit, wherein the merging operation is used to merge the target spatial unit with the adjacent units of the target spatial unit; The determination results of whether to perform a merging operation are determined by determining whether to perform a merging operation for the other spatial units among the plurality of spatial units besides the target spatial unit; Based on the determination results of whether the merging operation is performed for each of the multiple spatial units, multiple candidate path regions of the robot are determined.

7. The method according to any one of claims 1 to 6, characterized in that, Before obtaining the obstacle hardness parameters for the robot, the following steps are also included: Determine the force data of the robot within the target time period; Based on the force data, force change parameters corresponding to multiple sub-time periods are determined, wherein the target time period includes multiple sub-time periods; Based on the force change parameters corresponding to the multiple sub-time periods, determine the force intensity parameters corresponding to the robot; Based on the force intensity parameters, the obstacle hardness parameters corresponding to the robot are determined.

8. A device for determining the motion control strategy of a robot, characterized in that, include: The acquisition module is used to acquire the robot's obstacle space parameters, obstacle hardness parameters, and joint motion parameters; The first determining module is used to determine multiple candidate path regions of the robot based on the obstacle space parameters and the obstacle hardness parameters. The second determining module is used to determine motion risk parameters corresponding to the plurality of candidate path regions respectively, wherein the corresponding motion risk parameters represent the degree of risk of the robot moving in the corresponding candidate path region; The third determining module is used to determine stability constraint parameters corresponding to the plurality of candidate path regions based on the joint motion parameters and the obstacle stiffness parameters, wherein the corresponding stability constraint parameters are used to represent the degree of constraint of the corresponding candidate path region on the motion posture stability of the robot. The fourth determining module is used to determine the target path region from the multiple candidate path regions based on the motion risk parameters and stability constraint parameters corresponding to the multiple candidate path regions respectively; The fifth determining module is used to determine the robot's motion control strategy based on the motion risk parameters of the target path area.

9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the motion control strategy determination method for a robot as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the motion control strategy determination method for the robot as described in any one of claims 1 to 7.

Images