CN-122018540-A - Dynamic obstacle avoidance motion planning method for inspection robot

CN122018540ACN 122018540 ACN122018540 ACN 122018540ACN-122018540-A

Abstract

The invention relates to a dynamic obstacle avoidance motion planning method for a patrol robot, which belongs to the technical field of autonomous navigation and motion planning of robots and comprises the steps of acquiring a joint observation state of the patrol robot in a patrol environment; the method comprises the steps of extracting interaction relation features between a static obstacle and a dynamic obstacle and between a robot and the obstacle, constructing an Actor-Critic reinforcement learning frame, taking the extracted interaction relation features as input, outputting a robot motion strategy through a strategy network, evaluating state value through a value network, judging whether the relative speed of the robot and the dynamic obstacle is in an absolute collision area or not by a fusion speed obstacle method, eliminating collision risk speed, and outputting a robot linear speed and angular speed instruction according to the optimized robot motion strategy to realize autonomous obstacle avoidance motion in a dynamic environment. The invention realizes the autonomous obstacle avoidance navigation of the robot in a complex environment containing static and dynamic obstacles, which is safe and efficient and accords with social specifications.

Inventors

XU MINGKAI
PAN HUICHAO
LI CONG
LIU CHUNMING
WANG SIYUAN
WANG YUNPENG
WANG SHUYING

Assignees

国网山东省电力公司济南供电公司

Dates

Publication Date: 20260512
Application Date: 20251230

Claims (10)

1. The dynamic obstacle avoidance motion planning method for the inspection robot is characterized by comprising the following steps of: Step S1, acquiring the self state, the static obstacle state and the dynamic obstacle state of the inspection robot in the inspection environment through a plurality of sensors, and constructing a combined observation state; s2, extracting interaction relation features between static obstacles and dynamic obstacles and interaction relation features between robots and the obstacles based on a convolution social pooling network; S3, constructing an Actor-Critic reinforcement learning framework, taking the interaction relation features extracted in the step S2 as input, outputting a robot motion strategy through a strategy network, and evaluating state values through a value network; S4, a speed obstacle method is fused, whether the relative speed of the robot and the dynamic obstacle is in an absolute collision area or not is judged, and collision risk speed is eliminated, so that a robot motion strategy is optimized; And S5, outputting a robot linear speed and angular speed instruction according to the optimized robot motion strategy, and realizing autonomous obstacle avoidance motion in a dynamic environment.
2. The method for planning the dynamic obstacle avoidance movement of the inspection robot according to claim 1, wherein the step S1 comprises the steps of: Step S11, acquiring original data of a patrol environment through a laser radar, an inertial measurement unit and a vision sensor; Step S12, preprocessing the original data, wherein the preprocessing comprises time stamp alignment, coordinate system unification and noise filtering; step S13, a local coordinate system taking the robot direction as an x axis is established based on the current position of the robot; step S14, uniformly converting the state of the robot and the state of the obstacle into the local coordinate system; step S15, constructing a joint observation state vector containing robot state, obstacle state and time information.
3. The method for planning the dynamic obstacle avoidance motion of the inspection robot according to claim 2, wherein the state of the robot comprises a current position coordinate, a target position coordinate, a current linear velocity, a current angular velocity, a radius of a robot body and an attitude angle of the robot.
4. The method of claim 2, wherein the obstacle state comprises an obstacle type identifier, an obstacle center coordinate, an obstacle speed vector, an obstacle radius, and a relative distance and angle of the obstacle to the robot.
5. The method for planning the dynamic obstacle avoidance movement of the inspection robot according to claim 1, wherein the step S2 comprises the steps of: s21, designing a convolution social pooling network architecture, wherein the convolution social pooling network architecture comprises a coding module, a convolution social pooling module and a feature fusion module; Step S22, converting the variable number of obstacle states into a social feature representation with fixed dimension through an encoding module; S23, extracting spatial interaction relation features among barriers through a convolution social pooling module; s24, fusing the obstacle interaction characteristics with the state characteristics of the robot by a characteristic fusion module; step S25, outputting the comprehensive navigation feature vector containing the multi-level interaction relation.
6. The method for planning the dynamic obstacle avoidance movement of the inspection robot according to claim 5, wherein the step S3 comprises the steps of: s31, constructing an Actor-Critic reinforcement learning framework comprising a strategy network and a value network; Step S32, inputting the extracted comprehensive navigation features into a strategy network, and outputting the action probability distribution of the robot; Step S33, inputting the comprehensive navigation features into a value network, and outputting the value evaluation of the current state; step S34, adopting a near-end strategy optimization algorithm to perform joint training on the strategy network and the value network; step S35, training data are collected through interaction with the environment, and network parameters are continuously optimized.
7. The method for planning the dynamic obstacle avoidance movement of the inspection robot according to claim 1, wherein the step S4 comprises the steps of: Step S41, calculating a relative speed vector of the robot and each dynamic obstacle; step S42, constructing an absolute collision area based on a speed obstacle method, and judging whether the current relative speed is in the collision area or not; step S43, excluding speed options in the absolute collision area, and ensuring that the selected motion of the robot does not cause collision; step S44, introducing a relative speed evaluation item into the track evaluation function to encourage selection of a speed far away from the dynamic obstacle; And S45, dynamically adjusting the weight of the evaluation function according to the environmental risk, optimizing the motion strategy of the robot, and realizing the balance of safety and efficiency.
8. The method for planning dynamic obstacle avoidance movement of a patrol robot according to claim 7, wherein the trajectory evaluation function in step S44 is: , Wherein the method comprises the steps of As a function of the course angle evaluation, As an obstacle distance evaluation function, As a function of the speed evaluation, As a function of the relative velocity evaluation, Is a course angle evaluation function weight value, A weight value for evaluating the function of the obstacle, For the weight value of the speed evaluation function, For the weight of the relative velocity evaluation function, Indicating that normalization of each term is required.
9. The method for planning dynamic obstacle avoidance movement of a patrol robot according to any one of claims 1-8, wherein said step S5 comprises the steps of: Step S51, selecting an optimal action from the optimized action probability distribution, and acquiring corresponding linear velocity and angular velocity instructions; step S52, calculating the predicted pose of the robot at the next moment based on the differential kinematics model; Step S53, a control instruction is sent to a bottom layer motion controller to drive the robot to execute corresponding motion; step S54, repeating the steps S1-S4 in each control period to realize continuous dynamic obstacle avoidance; step S55, monitoring the execution result, if an abnormal situation occurs, starting a safety protection mechanism.
10. The method for planning dynamic obstacle avoidance movement of a patrol robot according to claim 9, wherein the differential kinematics model is expressed as: , Wherein, the For the position coordinates of the robot, Is the attitude angle of the robot, In order to be a line speed, In order to be able to achieve an angular velocity, For the control period.

Description

Dynamic obstacle avoidance motion planning method for inspection robot Technical Field The invention relates to a dynamic obstacle avoidance motion planning method for a patrol robot, and belongs to the technical field of autonomous navigation and motion planning of robots. Background The stable operation of the power grid core equipment is a key for guaranteeing the power supply and the national economy development. With the rapid increase of the power consumption of the whole society, the load pressure of power grid equipment is continuously increased, and urgent demands are put forward for regular and efficient inspection. The intelligent inspection robot of the transformer substation gradually replaces the traditional manual inspection. However, the existing substation inspection robot still has defects in the aspects of path planning accuracy, dynamic environment adaptability, autonomous obstacle avoidance capability and the like. Particularly, when the field environment changes and static and dynamic obstacles (such as staff) exist in the inspection process, the robot is difficult to quickly and safely adjust the inspection route. The traditional obstacle avoidance method (such as an artificial potential field method, a fuzzy control algorithm and the like) is generally based on a simple physical model or rule, so that complex interaction behaviors between dynamic obstacles are difficult to accurately predict and cope with, and the problems that a robot is in local optimum, a planned path is lengthy or is too close to the obstacle are easily caused. In recent years, reinforcement learning is introduced into the field of mobile robot navigation, but the existing method focuses on unilateral influence of modeling obstacles on robots, ignores complex social interaction relations between dynamic obstacles and between robots and obstacle groups, and has the problems of navigation safety and efficiency in a dense dynamic environment and meeting social standardization. Therefore, a method for planning the movement of the inspection robot, which can deeply understand the environment interaction relationship and make efficient safe obstacle avoidance decisions in real time, is needed. Disclosure of Invention In order to solve the problems, the invention provides a dynamic obstacle avoidance motion planning method for a patrol robot, which deeply fuses deep learning-based interactive relation understanding, reinforcement learning decision and classical collision speed elimination mechanism, and can realize safe, efficient and autonomous obstacle avoidance navigation conforming to social specifications of the robot in a complex environment containing static and dynamic obstacles. The technical scheme adopted by the invention for solving the technical problems is as follows: the embodiment of the invention provides a dynamic obstacle avoidance motion planning method for a patrol robot, which comprises the following steps: Step S1, acquiring the self state, the static obstacle state and the dynamic obstacle state of the inspection robot in the inspection environment through a plurality of sensors, and constructing a combined observation state; S2, extracting interaction relation characteristics between static obstacles and dynamic obstacles and interaction relation characteristics between robots and the obstacles based on a convolution social pooling network, wherein the convolution social pooling network comprises a coding module, a convolution social pooling module and a characteristic fusion module, the coding module divides a surrounding area of the robots into M multiplied by N regular grids, state vectors of each obstacle are mapped into characteristic vectors through multi-layer perceptors sharing parameters, and the characteristic vectors are filled into corresponding grids according to grid positions of the obstacles to generate a social characteristic diagram; S3, constructing an Actor-Critic reinforcement learning framework, taking the interaction relation features extracted in the step S2 as input, outputting a robot motion strategy through a strategy network, and evaluating state value through a value network, wherein the strategy network adopts a discrete action space design, and the action space comprises a plurality of groups of predefined linear velocity and angular velocity combinations; S4, a speed obstacle method is fused, whether the relative speed of the robot and the dynamic obstacle is in an absolute collision area or not is judged, and collision risk speed is eliminated, so that a robot motion strategy is optimized; And S5, outputting a robot linear speed and angular speed instruction according to the optimized robot motion strategy, and realizing autonomous obstacle avoidance motion in a dynamic environment. As a possible implementation manner of this embodiment, the step S1 includes the following steps: Step S11, acquiring original data of a patrol environment through a laser radar, an i