An obstacle avoidance method, medium, device and product for a water robot

Abstract

The application discloses an obstacle avoidance method, medium, equipment and product of a water robot, relates to the technical field of autonomous navigation of a water robot, and comprises the following steps: performing space-time synchronous fusion on multimodal environment data collected by a multisensor module of the water robot to obtain fused data; performing obstacle instance segmentation on the fused data, and classifying and state estimating the obstacles; defining an obstacle risk field according to the classification and state estimation of the obstacles; generating an initial collision-free path based on a static obstacle risk field in the obstacle risk field; constructing a smooth local path that satisfies a dynamic constraint and minimizes total cost and risk, and solving the local path within the corridor constraint of the initial collision-free path; and converting a point sequence of the local path into a smooth speed instruction executable by the robot, and performing fault-tolerant control based on the allowable speed. The application introduces a risk field model, and combines fault-tolerant control, so that the ability of the system to deal with unexpected conditions is improved.

Description

Technical Field

[0001] This invention relates to the field of autonomous navigation technology for aquatic robots, and in particular to a method, medium, device and product for obstacle avoidance of aquatic robots. Background Technology

[0002] With the development of automation technology, unmanned surface vehicles (USVs) have been widely used in various fields such as aquatic environment monitoring, hydrological surveys, and surface rescue. When performing tasks such as fixed-point water quality sampling or emergency rescue, USVs need to have a high level of autonomous navigation and obstacle avoidance capabilities. However, the aquatic environment is complex and changeable, with diverse types of obstacles, including static obstacles (such as reefs, fixed buoys, and bridge piers) and dynamic obstacles (such as passing vessels, swimmers, and floating objects). In addition, the water surface has interference such as waves and light reflections, which pose a severe challenge to the perception system.

[0003] Currently, common obstacle avoidance methods for aquatic robots have the following shortcomings:

[0004] (1) Limited perception methods and insufficient reliability: Most methods rely on a single sensor, such as using only a visual camera or only a lidar. Visual perception performance drops sharply in poor lighting and foggy weather, and it is difficult to measure distance accurately; while lidar has a large number of invalid points on the water surface (specular reflection) and has weak recognition ability for non-rigid obstacles (such as aquatic plants and foam). A single sensor cannot provide reliable obstacle information for all weather conditions and all types of obstacles.

[0005] (2) Obstacle handling is simplistic and lacks intelligent classification and risk assessment: Existing methods usually treat all detected point clouds or pixel clusters as obstacles without distinguishing their type (such as ships, floating objects, and shorelines) and motion state. This leads to a single obstacle avoidance strategy, which may overreact to harmless floating objects or under-assess the risks of high-speed ships, resulting in the planning of dangerous paths.

[0006] (3) Low coupling between path planning and task: Most obstacle avoidance path planning algorithms (such as A) DWA (Dynamic Path Avoidance) primarily considers geometric avoidance, without fully taking into account the impact of higher-level tasks (such as the necessity of reaching sampling points and the urgency of rescue) on the path. When an emergency rescue mission is triggered, it cannot make an optimal trade-off between safety and timeliness.

[0007] (4) Weak ability to predict dynamic obstacles: For dynamic obstacles such as moving ships, the simple speed obstacle method (VO) assumes that the speed is constant and does not consider the possibility of maneuvering and changing direction, resulting in inaccurate trajectory prediction and easy collision or planning oscillation. Summary of the Invention

[0008] The purpose of this invention is to address the lack of risk assessment in existing obstacle avoidance methods for aquatic robots, and to propose an obstacle avoidance method for aquatic robots, comprising the following steps: S1. Use the multi-sensor module on the water robot to collect multimodal environmental data, and perform spatiotemporal synchronous fusion of the multimodal environmental data to obtain fused data; S2. Perform obstacle instance segmentation on the fused data, and classify and estimate the state of the obstacles; S3. Define the obstacle risk field based on the category and state estimation of obstacles; S4. Based on the static obstacle risk field in the obstacle risk field, generate an initial collision-free path; construct a smooth local path that satisfies the dynamic constraints and minimizes the total cost and risk, and solve the local path within the corridor constraints of the initial collision-free path; S5. Convert the point sequence of the local path into a smooth speed command that the robot can execute, and perform fault-tolerant control based on the allowable speed.

[0009] Furthermore, the multi-sensor module includes a vision camera, a multi-line lidar, and a positioning and attitude determination system composed of an inertial measurement unit and a global navigation satellite system.

[0010] Furthermore, spatiotemporal synchronous fusion of multimodal environmental data is performed, specifically as follows: Sensor calibration is performed to obtain the extrinsic parameter matrix of the transformation from lidar to camera. Time synchronization of lidar and camera data is achieved using timestamps. Motion compensation of lidar and camera data is performed using inertial measurement unit data. The lidar point cloud data is projected onto the image pixel coordinate system using the following formula:

[0011] in, The homogeneous coordinates of the image, For the camera intrinsic parameter matrix, The extrinsic parameter matrix for the transformation of lidar to camera. The coordinates of the laser point; The semantic information of the image is back-projected onto the LiDAR point cloud, and each 3D point is assigned a preliminary semantic label.

[0012] Furthermore, S2 specifically refers to: S21. Divide the fused data into different obstacle instances; S22. Extract multi-dimensional feature vectors of obstacle instances, including: geometric features, semantic features, and dynamic features; S23. Input the multi-dimensional feature vector into a lightweight multi-layer neural network classifier and output the category of the obstacle and its confidence level. S24. Filter and predict the position and velocity of dynamic obstacles.

[0013] Furthermore, the obstacle risk field is defined as:

[0014] in, Represents the obstacle pair at time t Risk value at point This represents the base risk coefficient for the i-th obstacle category. This represents the i-th obstacle category. This represents the predicted position of the i-th obstacle at time t. These are parameters used to control the extent of risk spread. These are the weighting coefficients. It is the obstacle relative to the point relative velocity, That is the maximum expected speed. It is the trace of the covariance predicted by the Kalman filter. express Uncertainty covariance; The total risk field of the environment is the superposition of the risk fields of all obstacles.

[0015] Furthermore, the objective function for local path optimization is:

[0016]

[0017]

[0018]

[0019] in, The objective function for local path optimization is... Indicates a local path. Indicates task cost, This represents the cumulative sum of risks at path points. Represents the path curvature smoothing term. This represents a dynamic constraint term that constrains the changes in steering angle and velocity between path points within the robot's dynamic limits. Represents the k-th point of the local path The total risk field of the environment at any given moment. Represents the k-th point of the local path At time i, the obstacle pair Risk value at point This represents the (k+1)th point on the local path. This represents the k-th point on the local path. This represents the (k-1)th point on the local path. , , and These represent the weight coefficients for each item.

[0020] Furthermore, S6 specifically refers to: Fit a sequence of points along a local path to generate a parameterized continuous trajectory, and calculate the curvature of each point on the trajectory. Check the minimum rotation radius Is it greater than the robot's minimum rotation radius? , and Given the robot's maximum linear velocity and maximum angular velocity, if satisfied, based on the robot's centripetal acceleration limit. The allowable velocity curve is calculated based on the curvature at each point. The maximum permissible speed is taken, and finally a speed control command related to time t is generated.

[0021] The present invention also proposes a computer-readable storage medium storing a computer program, characterized in that: when the computer program is executed by a processor, it implements the above-described method.

[0022] The present invention also proposes an electronic device, characterized in that it includes a processor and a memory, the processor being interconnected with the memory, wherein the memory is used to store a computer program, the computer program including computer-readable instructions, and the processor is configured to invoke the computer-readable instructions to execute the above-described method.

[0023] The present invention also proposes a computer program product, including a computer program / instruction, characterized in that the computer program / instruction implements the above-described method when executed by a processor.

[0024] The beneficial effects of the technical solution provided by this invention are: (1) By deeply integrating vision and lidar, this invention makes full use of the rich semantic information of vision and the precise three-dimensional geometric information of lidar, overcomes the limitations of a single sensor in complex water environment, and achieves stable and accurate perception of diverse obstacles.

[0025] (2) The method provided by this invention not only detects obstacles, but also classifies and dynamically predicts them, and performs risk assessment in conjunction with the task context. This makes obstacle avoidance decision-making no longer a simple matter of avoidance, but rather an intelligent obstacle avoidance based on risk.

[0026] (3) By introducing a risk field model based on predictive uncertainty, the potential threat of dynamic obstacles can be assessed in a forward-looking manner. Combined with fault-tolerant control, this further enhances the system's ability to cope with emergencies. Attached Figure Description

[0027] Figure 1 This is a flowchart of an obstacle avoidance method for an aquatic robot according to an example of the present invention; Figure 2 This is a block diagram of an electronic device according to an exemplary embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0029] A flowchart of an obstacle avoidance method for an aquatic robot according to an example of the present invention is shown below. Figure 1 Specifically, it includes: S1. Use the multi-sensor module on the water robot to collect multimodal environmental data, and perform spatiotemporal synchronous fusion of the multimodal environmental data to obtain fused data.

[0030] Specifically, the multi-sensor module includes at least: a forward-looking panoramic vision camera, a multi-line lidar, and a positioning and attitude determination system consisting of an inertial measurement unit and a global navigation satellite system.

[0031] First, sensor calibration is performed to obtain the extrinsic parameter matrix between the camera and the LiDAR. (Transformation from lidar to camera). Time synchronization and motion compensation between camera and lidar data are achieved using hardware timestamps and inertial measurement unit data. For each lidar point... Projected onto the image pixel coordinate system using the following formula:

[0032] in, The homogeneous coordinates of the image, For the camera intrinsic parameter matrix, The extrinsic parameter matrix for the transformation of lidar to camera. These are the coordinates of the laser point.

[0033] Simultaneously, the image semantic information (pixel-level labels obtained through a pre-trained deep learning model) is back-projected onto the LiDAR point cloud, assigning preliminary semantic labels (such as "water", "boat", "floating object", "shore") to each 3D point.

[0034] S2. Perform obstacle instance segmentation on the fused data, and classify and estimate the state of the obstacles.

[0035] In this embodiment of the invention, a clustering algorithm based on Euclidean distance and semantic consistency (such as the improved DBSCAN algorithm) is used to segment the fused data into different obstacle instances. A multi-dimensional feature vector is extracted for each instance. The multi-dimensional feature vector includes: (1) geometric features, mainly bounding box size, point cloud density, and height distribution; (2) semantic features, mainly from the dominant category probability in vision; and (3) dynamic feature patterns, mainly the centroid velocity and angular velocity (if observable) calculated through the correlation of point clouds in consecutive frames. The multi-dimensional feature vector is input into a lightweight multi-layer neural network classifier, which outputs the fine-grained category of the obstacle. (e.g., large merchant ships, small fishing boats, stationary reefs, driftwood, clumps of seaweed, people who have fallen into the water) and their confidence levels. For dynamic obstacles, Kalman filtering or extended Kalman filtering is used to filter and predict position and velocity. The state equation and observation equation simplify to:

[0036]

[0037] in, This represents the state (including position and velocity) of the i-th dynamic obstacle at time k+1. This is the state transition matrix (assuming uniform or uniform angular velocity motion). For the observation matrix, , These are the covariance matrices of process noise and observation noise, respectively. Let i be the state of the i-th dynamic obstacle at time k. Let k be the noise of the control system at time k. The noise measured at time k, It is the measurement vector of the i-th dynamic obstacle at time k. This represents a Gaussian distribution with mean 0 and variance Q. This represents a Gaussian distribution with a mean of 0 and a variance of R.

[0038] The predicted future location of the obstacle and its uncertainty covariance can be obtained.

[0039] S3. Define the obstacle risk field based on the category and state estimation of the obstacles.

[0040] To quantify the threat level of each obstacle to the aquatic robot, a time-varying risk field is constructed, defined as a function of obstacle category, distance, relative velocity, and motion uncertainty. For each obstacle, its position at point... The risk value generated at this location is defined as:

[0041] in, Represents the obstacle pair at time t Risk value at point This represents the base risk coefficient for the i-th obstacle category. This represents the i-th obstacle category. This represents the predicted position of the i-th obstacle at time t. These are parameters used to control the extent of risk spread. This is a weighting factor; its purpose is to increase the risk of obstacles approaching at high speeds. It is the obstacle relative to the point relative velocity, That is the maximum expected speed. It is the trace of the Kalman filter prediction covariance, representing the uncertainty in motion prediction. The greater the uncertainty, the more diffuse the risk field. express Uncertainty covariance, It is a Gaussian distance decay term centered on the predicted obstacle location. It is the speed term.

[0042] The basic risk coefficient can be set as follows: people falling into the water > high-speed vessels > stationary reefs > floating objects.

[0043] Total environmental risk field It is the superposition of all obstacle risk fields.

[0044]

[0045] S4. Based on the static obstacle risk field in the obstacle risk field, generate an initial collision-free path, construct a smooth local path that satisfies the dynamic constraints and minimizes the total cost and risk, and solve the local path within the corridor constraints of the initial collision-free path.

[0046] In this embodiment of the invention, local path replanning is performed according to the current mission mode (cruise monitoring mode or emergency rescue mode).

[0047] Current position and target point The path planning problem between (the next sampling point or the location of the person who fell into the water) is modeled as an optimization problem. A hierarchical optimization framework is adopted, as follows: 1) Global Guidance Layer: In a simplified risk field considering static obstacles, A is used. The algorithm or Quick Random Tree (RRT) generates an initial coarse collision-free path. .

[0048] 2) Local optimization layer: In Within the corridor constraints, a gradient descent-based optimization method is used to find a smooth local path that satisfies the dynamic constraints and minimizes the total cost and risk. Let n be the number of points in the local path. The objective function for optimization is... for:

[0049]

[0050]

[0051]

[0052] in: Indicates task cost, Below, straight paths are encouraged to save energy and ensure sampling quality; Below, for The emphasis was on getting as close to the goal as possible. This represents the cumulative risk at each path point. This represents the path curvature smoothing term. This represents a dynamic constraint term that constrains the changes in steering angle and velocity between path points within the robot's dynamic limits. The weighting coefficients for each sub-item are dynamically adjusted under different task modes, for example, Down, Significantly increased, The risk is relatively reduced, allowing robots to take on a certain degree of risk for the sake of timeliness (but they still need to avoid extremely high-risk areas).

[0053] S5. Convert the point sequence of the local path into a smooth speed command that the robot can execute, and perform fault-tolerant control based on the allowable speed.

[0054] In this embodiment of the invention, B-spline curves or polynomial curves are used to... A fitting process is performed to generate a parameterized continuous trajectory. The curvature at each point on the trajectory is then calculated. Maximum curvature Corresponding minimum turning radius ,examine Is it greater than the robot's minimum turning radius (by...) and Decide, If satisfied, according to the robot's centripetal acceleration limit. Calculate the permissible speed curve The maximum permissible speed is taken, and finally, a speed control command related to time t is generated. Send it to the underlying propulsion controller.

[0055] Simultaneously, a safety monitor is set up to continuously compare the predicted risk field with actual sensor feedback. If an unexpected obstacle or a surge in risk is detected, rapid replanning in steps two through five is immediately triggered to ensure safety.

[0056] This embodiment uses a water robot as a platform to perform the task of "starting from the dock to conduct water quality patrol sampling, receiving an alarm from a person who has fallen into the water, and turning to rescue them".

[0057] (1) Platform and initial conditions: Robot's initial pose: m, heading angle (Due east).

[0058] Cruise mission target point (sampling point 1): m.

[0059] We received an emergency call reporting a person who had fallen into the water. Location: m.

[0060] Robot's maximum speed: m / s, maximum steering angular velocity: rad / s.

[0061] Sensors: 16-line lidar (360° horizontal FOV, maximum range 50m), 1080p forward-looking camera.

[0062] (2) Scene setting: When the robot goes to The following obstacles were detected during the journey: A small, stationary fishing boat, category =“Stationary ship”, position m, basic risk coefficient .

[0063] A piece of floating wood debris, category =“Floating object”, location m, basic risk coefficient .

[0064] A yacht turning right (north), initial detection position. m, velocity m / s, category ="Motorized Vessels", Basic Risk Factor .

[0065] Person who fell into the water (added to the risk area after alarm is triggered), location m, velocity approximately 0, category =“People who have fallen into the water”, basic risk level (Highest priority, but as a target that needs to be approached, its risk field function is specially treated and is reflected as attractiveness in the path cost).

[0066] (3) Implementation process and numerical calculation: Step 1: Multimodal sensor data acquisition and spatiotemporal synchronization fusion.

[0067] Assuming that through fusion and classification, the following can be successfully identified: For dynamic obstacles Perform KF tracking.

[0068] Step 2: Obstacle instance segmentation, classification, and state estimation based on fused data.

[0069] Assume process noise Observation noise After filtering, at the planned time... ,get estimated state Predicting covariance traces .

[0070] Step 3: Modeling dynamic obstacle threat assessment based on risk field.

[0071] Setting parameters: m, , m / s. To assess the robot's current position. by Take threats as an example. Prediction. exist Location at any moment Relative velocity exist When calculating a point, it is assumed that the point is stationary. , module length m / s. Calculation exist Risk value of the point:

[0072] It is clear that due to the great distance, the risk is almost zero. However... Nearby, for example, a point At a distance of approximately 5.2 meters, the risk level will increase dramatically.

[0073] Step 4: Task-driven hierarchical local path planning and optimization.

[0074] Phase 1: Cruise Monitoring Mode (Go to...) The weights are set as follows: (Tends to be flat) , , Global boot path Possibly from Left side and It passes between the right sides. After local optimization, a cruise path is generated. The path will smoothly move away from... and and to Be wary of its movement direction (northward) and consider entering slightly later to avoid its predicted path. Assume the optimized pathpoint sequence contains... wait.

[0075] Phase Two: Emergency Rescue Mode Triggered (Target Switched to) Upon receiving an alarm, the task mode immediately switches to emergency rescue mode. The weights are dynamically adjusted as follows: (Emphasizing rapid approach) (Safety is still a priority, but tolerance has increased.) , The current robot position assumption has been established. .

[0076] Using a new global guidance approach, the goal becomes A The algorithm considers static obstacles and At a fixed position, generate a possible path from South side and The initial path that crosses between the north sides.

[0077] Further local optimization is adopted, using an optimizer to perform calculations... At that time, one will strongly feel from The dynamic risk field. Due to It is moving northward, and its risk field is expanding to the north. The optimizer will attempt to find... (Hoping to get a straight line closer to the person who fell into the water) and The balance between them. We use a simplified discrete-point evaluation to illustrate the decision: Suppose there are two choices for the candidate target point at the next time step: Point A: (More directly pointing to the person who fell into the water, but far from the prediction) (The path is relatively short). Point B: (Slightly north, further away) (Predicted path). Calculate from the current position. The vectors to points A and B. Assume a simple risk assessment is based on the midpoint of the path. and The risk value (calculated using the formula in step S3, requiring specific numerical values ​​to be substituted). Simultaneously, calculate the task cost, i.e., the cost to reach the final goal. The remaining distance.

[0078] Therefore, a rough calculation yields the following: Path A: Smaller (because the direction is more direct), but Larger (midpoint distance) close).

[0079] Path B: Slightly larger (detour), but Smaller. Under the weight of emergency rescue mode ( The optimizer might choose point A because task timeliness has a higher weight, and the calculated risk hasn't exceeded the absolute threshold. However, optimization is continuous, and ultimately... It will be a smooth curve, subtly adjusting to minimize its influence while rapidly approaching the overall trend of the person falling into the water. Close contact.

[0080] Step 5: Smooth trajectory generation and fault-tolerant control execution.

[0081] right Perform B-spline fitting. Assume a segment is generated from... arrive Calculate the curvature of the curve. Maximum curvature. Corresponding minimum turning radius ,examine Is it greater than the robot's minimum turning radius (by...) and Decide, If satisfied, then generate the velocity according to the curvature constraint (m). instruction.

[0082] During the cruising phase, the robot smoothly avoided stationary fishing boats and floating debris, and made preventative avoidance maneuvers against motorized vessels. Upon receiving a rescue order, the robot quickly replanned its path, and with the support of dynamic risk assessment, selected the fastest route to reach the person in the water within a safe margin, successfully avoiding a turning yacht. It arrived at the rescue site in a timely manner. Throughout the process, the obstacle avoidance behavior was highly coordinated with the mission objective, demonstrating the intelligence and practicality of this invention.

[0083] In one exemplary embodiment, a computer-readable storage medium is included, which stores a computer program that, when executed by a processor, implements the method described above.

[0084] Please see Figure 2In one exemplary embodiment, the device further includes an electronic device including at least one processor, at least one memory, and at least one communication bus.

[0085] The memory stores a computer program, which includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through a communication bus to execute the above method.

[0086] In one exemplary embodiment, a computer program product is proposed, including a computer program / instructions that, when executed by a processor, implement the method described above.

[0087] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for obstacle avoidance by an aquatic robot, characterized in that, Includes the following steps: S1. Use the multi-sensor module on the water robot to collect multimodal environmental data, and perform spatiotemporal synchronous fusion of the multimodal environmental data to obtain fused data; S2. Perform obstacle instance segmentation on the fused data, and classify and estimate the state of the obstacles; S3. Define the obstacle risk field based on the category and state estimation of obstacles; S4. Based on the static obstacle risk field in the obstacle risk field, generate an initial collision-free path; construct a smooth local path that satisfies the dynamic constraints and minimizes the total cost and risk, and solve the local path within the corridor constraints of the initial collision-free path; S5. Convert the point sequence of the local path into a smooth speed command that the robot can execute, and perform fault-tolerant control based on the allowable speed.

2. The obstacle avoidance method for an aquatic robot according to claim 1, characterized in that, The multi-sensor module includes a vision camera, a multi-line lidar, and a positioning and attitude determination system that combines an inertial measurement unit and a global navigation satellite system.

3. The obstacle avoidance method for an aquatic robot according to claim 2, characterized in that, Spatiotemporal synchronous fusion of multimodal environmental data is performed, specifically as follows: Sensor calibration is performed to obtain the extrinsic parameter matrix of the transformation from lidar to camera. Time synchronization of lidar and camera data is achieved using timestamps. Motion compensation of lidar and camera data is performed using inertial measurement unit data. The lidar point cloud data is projected onto the image pixel coordinate system using the following formula: in, The homogeneous coordinates of the image, For the camera intrinsic parameter matrix, The extrinsic parameter matrix for the transformation of laser radar to camera. The coordinates of the laser point; The semantic information of the image is back-projected onto the LiDAR point cloud, and each 3D point is assigned a preliminary semantic label.

4. The obstacle avoidance method for an aquatic robot according to claim 1, characterized in that, S2 specifically refers to: S21. Divide the fused data into different obstacle instances; S22. Extract multi-dimensional feature vectors of obstacle instances, including: geometric features, semantic features, and dynamic features; S23. Input the multi-dimensional feature vector into a lightweight multi-layer neural network classifier and output the category of the obstacle and its confidence level. S24. Filter and predict the position and velocity of dynamic obstacles.

5. The obstacle avoidance method for an aquatic robot according to claim 1, characterized in that the obstacle... The risk field is defined as: in, Represents the obstacle pair at time t Risk value at point This represents the base risk coefficient for the i-th obstacle category. This represents the i-th obstacle category. This represents the predicted position of the i-th obstacle at time t. It is a parameter used to control the scope of risk diffusion. These are the weighting coefficients. It is the obstacle relative to the point relative velocity, That is the maximum expected speed. It is the trace of the covariance predicted by the Kalman filter. express Uncertainty covariance; The total risk field of the environment is the superposition of the risk fields of all obstacles.

6. The obstacle avoidance method for an aquatic robot according to claim 5, characterized in that, The objective function for local path optimization is: in, The objective function for local path optimization is... Indicates a local path. Indicates task cost, This represents the cumulative sum of risks at path points. Represents the path curvature smoothing term. This represents a dynamic constraint term that constrains the changes in steering angle and velocity between path points within the robot's dynamic limits. Represents the k-th point of the local path The total risk field of the environment at any given moment. Represents the k-th point of the local path At time i, the obstacle pair Risk value at point This represents the (k+1)th point on the local path. This represents the k-th point on the local path. This represents the (k-1)th point on the local path. , , and These represent the weight coefficients for each item.

7. The obstacle avoidance method for an aquatic robot according to claim 6, characterized in that, S6 specifically refers to: Fit a sequence of points along a local path to generate a parameterized continuous trajectory, and calculate the curvature of each point on the trajectory. Check the minimum rotation radius Is it greater than the robot's minimum rotation radius? , and Given the robot's maximum linear velocity and maximum angular velocity, if satisfied, based on the robot's centripetal acceleration limit. The allowable velocity curve is calculated based on the curvature at each point. The maximum permissible speed is taken, and finally a speed control command related to time t is generated.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, The device includes a processor and a memory, the processor being interconnected with the memory, wherein the memory is used to store a computer program, the computer program including computer-readable instructions, and the processor is configured to invoke the computer-readable instructions to perform the method as described in any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the method described in any one of claims 1 to 7.

Images