Self-adaptive obstacle avoidance method for cooperative mechanical arm in dynamic scene

Abstract

The invention provides a cooperative mechanical arm self-adaptive obstacle avoidance method in a dynamic scene, which comprises the following steps of: 1, setting an initial pose and a target pose according to a work task of a cooperative robot, and performing global path planning in an initial environment; 2, an operator completes an assembly task in an assembly area, and the collaborative robot operates according to a preset route and completes the assembly task in a matched mode; 3, the depth camera obtains a human body skeleton and skeleton node information, and a human body bounding box is generated; 4, whether the distance between the arm bounding box and the robot is smaller than the safety distance or not is judged, and if yes, collision early warning is triggered; 5, a new collision-free track is re-planned based on the self-adaptive obstacle avoidance strategy of the artificial potential field method, and the mechanical arm moves according to the new track to avoid the obstacle; and 6, after the obstacle leaves the safety early warning range, the mechanical arm returns to the global path and finally arrives at the target point.

Description

Technical Field

[0001] The present invention belongs to the technical field of obstacle avoidance in human-machine collaborative scenarios, and specifically relates to an adaptive obstacle avoidance method for collaborative robotic arms in dynamic scenarios. Background Art

[0002] In complex assembly environments, intelligent robots need to have intelligent perception, decision-making, and action capabilities to adapt to dynamically changing scenarios, such as avoiding moving obstacles and responding quickly. Dynamic obstacle avoidance technology has become an important issue in intelligent robot research.

[0003] In the field of customized and small-batch industrial production, collaborative robotic arms are widely used. As the degree of human-machine collaboration continues to improve, the interaction between humans and robotic arms becomes closer and closer. In order to ensure the safety of humans and collaborative robotic arms, active obstacle avoidance strategies are combined before a collision occurs to avoid operators, thereby preventing collisions or collaborative conflicts. Therefore, the adaptive obstacle avoidance strategy of the robotic arm is of great significance to ensure human-machine safety. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide an adaptive obstacle avoidance method for collaborative robotic arms in dynamic scenarios, so as to solve the problem that the storage capacity of various types of existing garbage is unevenly distributed and the garbage storage device cannot automatically adjust the garbage capacity.

[0005] In order to achieve the above technical features, the purpose of the present invention is achieved as follows: a method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene, comprising the following steps:

[0006] Step 1: Set the initial pose and target pose according to the collaborative robot’s work task, and perform global path planning in the initial environment;

[0007] Step 2: The operator completes the assembly task in the assembly area, and the collaborative robot runs according to the predetermined route to complete the assembly task;

[0008] Step 3: The depth camera obtains the human skeleton and skeleton node information and generates a human body bounding box;

[0009] Step 4: Determine whether the distance between the arm bounding box and the robot is less than the safe distance. If so, a collision warning is triggered.

[0010] Step 5: The adaptive obstacle avoidance strategy based on the artificial potential field method replans a new collision-free trajectory, and the robot moves along the new trajectory to avoid obstacles;

[0011] Step 6: After the obstacle leaves the safety warning range, the robot returns to the global path and finally reaches the target point.

[0012] The global path considers obstacles in the environment as static obstacles in the initial state, and uses the improved RRT algorithm to expand toward the target posture with the initial posture as the root node.

[0013] The improved RRT algorithm adopts a target bias dynamic sampling strategy, an adaptive step size strategy and path pruning optimization for the RRT algorithm.

[0014] The human skeleton and skeleton nodes are only composed of nodes of the upper body of the human body, and the human body bounding box also only generates a bounding box of the upper body.

[0015] The collision warning is achieved by simplifying the models of the human body and the collaborative robot arm and adopting the shortest distance collision warning method in combination with the human body motion prediction results.

[0016] The adaptive obstacle avoidance strategy based on the artificial potential field method proposes an adaptive point selection obstacle avoidance strategy, realizes local obstacle avoidance path replanning, designs a local minimum escape strategy based on plane constraint sampling, and combines piecewise polynomial interpolation to smooth the trajectory of the obstacle avoidance path.

[0017] The improved RRT algorithm is used to quickly generate a global path, and the cubic spline interpolation function is used in combination with the improved particle swarm algorithm to optimize the trajectory with the shortest time to generate a global trajectory.

[0018] The camera is used to perceive the human skeleton information in the environment, and the human skeleton time series information is predicted. Combined with hand-eye calibration, the skeleton point information is converted to the robot base coordinate system to perform human-machine shortest distance collision warning detection.

[0019] When the collision warning is not triggered, it runs according to the global trajectory. When the collision warning is triggered, the dangerous part of the original trajectory is eliminated, and the adaptive obstacle avoidance re-planning method is used to generate a new avoidance trajectory. The disconnected trajectories are smoothly connected in combination with the 3-5-3 piecewise polynomial and the new trajectory is input into the robot control system. Finally, the human skeleton information is continuously detected and updated through the intelligent perception system, and the collaborative robot dynamic obstacle avoidance is achieved through collision warning detection and trajectory re-planning, ensuring the safety of man and machine while completing the established work tasks.

[0020] The improved particle swarm algorithm dynamically adjusts the inertia weight and learning factor in the particle swarm algorithm to balance the global search and local search, promote the algorithm to explore the global area in the early stage and accurately search the local area in the later stage, ensure the diversity of the particle population, and thus improve the convergence speed and search efficiency.

[0021] The present invention has the following beneficial effects:

[0022] 1. The present invention proposes a global planning method based on an improved RRT algorithm in a static unstructured environment. In view of the randomness and blindness of the RRT algorithm sampling, a target bias dynamic sampling strategy is introduced to increase the exploration of the target area and improve the goal orientation of the algorithm. At the same time, in order to avoid slow local expansion, an adaptive step size strategy is introduced to adaptively adjust the step size according to the number of obstacles near the end effector of the manipulator, reduce the number of redundant nodes in the open area, improve the convergence speed of the algorithm, and shorten the path length by pruning and optimizing the path and removing redundant path points.

[0023] 2. The present invention performs rapid enveloping based on the bounding box model of the human skeleton and the collaborative robot arm, and proposes a collision warning method based on the shortest distance. Combined with the results of human motion prediction, the potential risks between the human body and the collaborative robot arm can be calculated in advance, thereby improving the efficiency of collision warning, enhancing the safety of human-machine collaboration, and realizing timely warning of potential collision risks, thereby improving the safety and efficiency of human-machine collaboration.

[0024] 3. The present invention proposes an adaptive obstacle avoidance replanning method based on the improved artificial potential field method. First, an adaptive point selection obstacle avoidance strategy is proposed to realize the local obstacle avoidance path replanning and ensure the distance between the robot arm and the dynamic obstacle. Secondly, a local minimum escape strategy based on plane constraint sampling is designed. Finally, the newly generated local obstacle avoidance trajectory and the global trajectory are smoothly connected in position, velocity and acceleration by combining the 3-5-3 piecewise polynomial interpolation function, thereby obtaining a continuous and collision-free obstacle avoidance trajectory. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0026] Figure 1 The figure is a schematic diagram of a specific implementation process of an embodiment of the present invention.

[0027] Figure 2 Schematic diagram of a simplified human upper body skeleton according to an embodiment of the present invention.

[0028] Figure 3 FIG. 4 is a schematic diagram of a human body envelope according to an embodiment of the present invention.

[0029] Figure 4 Schematic diagram of the adaptive point selection strategy according to an embodiment of the present invention.

[0030] Figure 5 Schematic diagram of a plane constraint sampling mechanism according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] In conjunction with the drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0032] Embodiment 1:

[0033] See also Figure 1-5 ,

[0034] A method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene, comprising the following steps:

[0035] Step 1: Set the initial pose and target pose according to the collaborative robot’s work task, and perform global path planning in the initial environment;

[0036] The global path is to regard obstacles in the environment as static obstacles in the initial state, and use the improved RRT algorithm to expand toward the target posture with the initial posture as the root node;

[0037] The improved RRT algorithm adopts a target bias dynamic sampling strategy, an adaptive step size strategy and path pruning optimization for the RRT algorithm.

[0038] Step 2: The operator completes the assembly task in the assembly area, and the collaborative robot runs according to the predetermined route to complete the assembly task;

[0039] Step 3: The depth camera obtains the human skeleton and skeleton node information and generates a human body bounding box;

[0040] The human skeleton and skeleton nodes are only composed of nodes of the upper body of the human body, and the human body bounding box also only generates a bounding box of the upper body.

[0041] Step 4: Determine whether the distance between the arm bounding box and the robot is less than the safe distance. If so, a collision warning is triggered.

[0042] The collision warning is achieved by simplifying the models of the human body and the collaborative robot arm and adopting the shortest distance collision warning method in combination with the human body motion prediction results.

[0043] Step 5: The adaptive obstacle avoidance strategy based on the artificial potential field method replans a new collision-free trajectory, and the robot moves along the new trajectory to avoid obstacles;

[0044] The adaptive obstacle avoidance strategy based on the artificial potential field method proposes an adaptive point selection obstacle avoidance strategy, realizes local obstacle avoidance path replanning, designs a local minimum escape strategy based on plane constraint sampling, and combines piecewise polynomial interpolation to smooth the trajectory of the obstacle avoidance path.

[0045] Step 6: After the obstacle leaves the safety warning range, the robot returns to the global path and finally reaches the target point.

[0046] Furthermore, an improved RRT algorithm is used to quickly generate a global path, and the cubic spline interpolation function is used in combination with an improved particle swarm algorithm to optimize the trajectory with the shortest time to generate a global trajectory.

[0047] Furthermore, a camera is used to perceive the human skeleton information in the environment, and the human skeleton time series information is predicted. Combined with hand-eye calibration, the skeleton point information is converted to the robot base coordinate system to perform human-machine shortest distance collision warning detection.

[0048] Furthermore, when the collision warning is not triggered, it runs according to the global trajectory. When the collision warning is triggered, the dangerous part of the original trajectory is eliminated, and the adaptive obstacle avoidance re-planning method is used to generate a new avoidance trajectory. The disconnected trajectories are smoothly connected in combination with the 3-5-3 piecewise polynomial and the new trajectory is input into the robot control system. Finally, the human skeleton information is continuously detected and updated through the intelligent perception system, and the collaborative robot dynamic obstacle avoidance is realized through collision warning detection and trajectory re-planning, ensuring the safety of man and machine while completing the established work tasks.

[0049] Furthermore, the improved particle swarm algorithm dynamically adjusts the inertia weight and the learning factor in the particle swarm algorithm to balance the global search and the local search, promotes the algorithm to explore the global area in the early stage and accurately search the local area in the later stage, ensures the diversity of the particle population, and thus improves the convergence speed and search efficiency.

[0050] Embodiment 2:

[0051] like Figure 1 The adaptive obstacle avoidance method of a collaborative robot arm in a dynamic scene includes the following parts:

[0052] In the global motion planning part, the initial and final postures of the collaborative robot and the initial static environment information are given according to the task requirements, and an improved RRT algorithm is used to quickly generate a global path. The cubic spline interpolation function is used in combination with the improved particle swarm algorithm to optimize the trajectory with the shortest time to generate a global trajectory.

[0053] In the collision warning part, the Kinect V2 camera is used to perceive the human skeleton information in the environment. The human skeleton time series information is predicted, and the skeleton point information is converted to the robot base coordinate system in combination with hand-eye calibration to perform human-machine shortest distance collision warning detection.

[0054] The dynamic obstacle avoidance system of the collaborative robot arm runs according to the global trajectory when the collision warning is not triggered. When the collision warning is triggered, the dangerous part of the original trajectory is eliminated, and an adaptive obstacle avoidance re-planning method is used to generate a new avoidance trajectory. The disconnected trajectories are smoothly connected by combining the 3-5-3 piecewise polynomial and the new trajectory is input into the robot control system. Finally, the intelligent perception system can continuously detect and update the human skeleton information, and realize dynamic obstacle avoidance of the collaborative robot through collision warning detection and trajectory re-planning, ensuring the safety of man and machine while completing the established work tasks.

[0055] The probability-based target bias strategy in this example is to set a probability threshold ε target ∈(0, 1), and randomly select a value ε between 0 and 1, when ε is not less than ε target When, in the collision-free space χ free Randomly select points inside; when ε is less than ε target When the target point is used as the sampling point, the expansion direction of the tree will be towards X goal Directional expansion. Generate a dynamically changing sampling area near the target point, so that the sampling area becomes a circular area with the target point as the center and a dynamically changing radius. When there are no obstacles, this area is feasible. The radius of the sampling circle should be reduced to guide the node to approach the target point; when encountering obstacles, the sampling radius should be expanded to find other feasible expansion nodes to avoid obstacles. By designing this strategy, the random tree has a certain target bias.

[0056] In this example, the adaptive step length strategy is to adaptively adjust the search step length according to the number of obstacles near the end effector of the collaborative robot arm. The adaptive step length calculation formula is as follows:

[0057]

[0058] Where n obs is the number of obstacles near the end effector of the collaborative robot arm, and w is the influence coefficient of the number of obstacles on the step length. When there are no obstacles near the end effector of the collaborative robot arm, the step length is 2S0. When the number of obstacles increases, the step length gradually decreases. This strategy can adjust the search step length more flexibly in different situations, reduce the number of iterations, and improve the algorithm's exploration ability in complex environments.

[0059] In this example, path pruning optimization is to prune redundant paths to improve path smoothness.start As the starting point, the key points (X start ,X1,χ2,···,χ goal ) to traverse and judge χ start Whether the line connecting the key points of the subsequent path collides with the obstacle, if not, then delete the key points of the path between the two nodes; if a collision occurs, then connect χ start and the previous node of the node, and take the previous node of the node as the new starting point, continue to traverse the subsequent nodes of the node, and repeat the above operations until the target point x is reached goal .

[0060] The global path generated by the improved RRT algorithm in this example is connected by discrete points. Before the global path is applied to the robot movement, the cubic spline interpolation function is used for trajectory planning. It has continuous first-order and second-order derivatives. This feature can ensure the continuity of displacement, velocity and acceleration of the collaborative robot during the task execution. The improved particle swarm algorithm is used to allocate the time for each trajectory. The particle swarm algorithm is as follows:

[0061]

[0062] In the formula, ω is the inertia weight, c1 is the individual self-learning factor, c2 is the group learning factor, r1 and r2 are random numbers between [0, 1], P id is the individual optimal particle position, P gd is the global optimal particle position. The values ​​of ω, c1, and c2 in the PSO algorithm will directly affect the convergence performance of the algorithm. In the traditional PSO algorithm, these three values ​​are constants, which easily cause the algorithm to fall into the local optimum. Therefore, this section uses trigonometric functions to adjust the inertia weight and learning factor, and constructs the ω, c1, and c2 value functions as follows:

[0063]

[0064] After the improvement, by dynamically adjusting the inertia weight and learning factor, the global search and local search can be balanced, which promotes the algorithm to explore the global area in the early stage and accurately search the local area in the later stage, ensuring the diversity of the particle population, thereby improving the convergence speed and search efficiency.

[0065] In this example, the Kinect V2 camera is used to perceive the human skeleton information in the environment and simplify the human skeleton as follows: Figure 2 The joints shown include the neck, left shoulder, right shoulder, left elbow, right elbow, left fingertips, right fingertips, and spine. Figure 2The joints 2, 4, 8, 5, 9, 21, 23, 0 in the figure are the upper body, the joint line 0-2 is the left arm, and the joint line (8-9-23) is the right arm. The coordinate changes of these 8 points (0, 2, 4, 5, 8, 9, 21, 23) can reflect the movement of the upper body. The three-dimensional coordinates of the key point u in the i frame are represented by the vector express:

[0066]

[0067] In the formula, u represents the eight key points of the upper body u=(0,2,4,5,8,9,21,23), and these eight key points constitute a skeleton feature J of the i-th frame i , such as:

[0068]

[0069] Then the time series of the upper body skeleton features from frame r to frame r+n can be expressed as f r,n =

[0070] [J r ,J r+1 ,…,J r+n ] T The skeleton time series is used as the input of the LSTM network, and the output is the trajectory sequence of the predicted 8 key points at subsequent times.

[0071] In this example, the human body and collaborative robot arm models are simplified by combining the bounding sphere and the cylinder to form a capsule-like bounding box model. First, for the human body, the human body is divided into a capsule combination model consisting of two arms (the left arm is connected by the 4-5-21 line, and the right arm is connected by the 8-9-23 line) and the torso (0-2 line). In this way, the human body model can be encapsulated by 5 capsule models, such as Figure 3 .

[0072] In this example, the collision warning method based on the shortest distance introduces a safety factor K on the original minimum safety distance formula. safe Minimum safety distance S min The calculation formula is as follows:

[0073]

[0074] D safe =K safe ·S min ;

[0075] In the formula, v R represents the component of the robot's velocity along the direction of the operator (mm / s), T Rrepresents the robot response time (s), v H represents the operator's component along the robot's motion direction (mm / s), T S and T R Represent the robot's braking time and response time (s), a s Represents the robot deceleration (mm / s 2 ), e represents the sensor measurement system accuracy (mm), Z R Represents the uncertainty of the robot's position, the safety factor K safe Usually a constant greater than 1 is selected, and the specific value can be adjusted according to the actual working conditions of the robot.

[0076] In this example, the dynamic obstacle avoidance of the collaborative robot arm retains the original useful path part, and on this basis plans an avoidance path that can return to the global path. It is a local path replanning based on the adaptive obstacle avoidance strategy switching mechanism of the artificial potential field method.

[0077] This strategy adaptively selects obstacle avoidance path points according to the potential field force and the influence range of obstacle repulsion, forming a local obstacle avoidance path such as Figure 4 As shown, the distance between the global path and the obstacle is selected to be greater than the obstacle repulsion influence range D safe The nodes of the robot are combined into a safe node set so that the robot can return to the original motion after successfully avoiding the obstacle. A point closest to the current robot position is selected from the safe node set as a temporary target point. The direction of the resultant force on the robot at this point is determined by the artificial potential field method. The path nodes are expanded along the direction of the resultant force. When the distance between the expanded node and the obstacle is greater than D safe , mark it as obstacle avoidance point 1, draw a perpendicular line from the obstacle position to the line between obstacle avoidance point 1 and the temporary target point, and extend a point on the extension line of the perpendicular line whose distance from the obstacle is greater than D safe The node is recorded as obstacle avoidance point 2, then the replanned path is composed of the current position of the robot end effector, obstacle avoidance point 1, obstacle avoidance point 2 and the temporary target point. On this basis, the idea of ​​sliding time window is introduced. The sliding window is refreshed once every time step of the robot movement, the environment and predicted position are updated, and it is judged whether the collision warning is triggered. When the collision warning is triggered, the robot replans the obstacle avoidance path in combination with the adaptive point selection obstacle avoidance strategy.

[0078] The plane constraint sampling mechanism of this example is a plane formed by the robot's current position, the target point, and the obstacle position. On this plane, with the robot's gravitational direction as the center, constraint sampling is performed at intervals of 30° within a range of 180° toward the target point. Multiple sampling points can be obtained, such as Figure 5As shown. Within the influence range of the obstacle, those sampling points that are closer to the obstacle will be subject to greater repulsion, while at the sampling points at the edge of the obstacle repulsion range, the repulsion is relatively small. At the same time, the gravitational force will be smaller at those locations that are closer to the target point. Considering that in actual tasks, the robot needs to be able to reach the target as quickly as possible while maintaining a certain distance from the obstacle. Therefore, it is necessary to select a point at the edge of the obstacle influence range and closer to the target point as the sampling direction. From the perspective of potential field force, it is necessary to balance repulsion and gravitation, that is, to select a sampling direction with less repulsion and gravitation as the escape direction. The cost function can be composed of repulsion and gravitation:

[0079] CostF=αF att (χ)+βF rep (χ);

[0080] In the formula, α is the penalty strength of gravity, and β is the penalty strength of repulsion. Calculate the cost function corresponding to each sampling point respectively, and select the minimum value as the direction of the robot's local extremum escape.

[0081] The path replanned by the adaptive obstacle avoidance strategy of this example includes the current position, obstacle avoidance point 1, obstacle avoidance point 2 and the temporary target point. Three smooth trajectories need to be constructed. This embodiment uses a 3-5-3 polynomial interpolation function for trajectory planning, that is, the starting segment uses a third-order polynomial interpolation, the middle segment uses a fifth-order polynomial interpolation, and the terminal segment uses a third-order polynomial interpolation. The 3-5-3 polynomial interpolation function is as follows:

[0082]

[0083] In the formula, θ i1 ,θ i2 ,θ i3 They represent the first, second, and third polynomial trajectory functions of the i-th joint of the collaborative robot arm, i = 1, 2, 3, ···, 6; a i1j 、a i2j 、a i3j They are the j-th coefficients of the 1st, 2nd, and 3rd segment polynomial interpolations of the i-th joint interpolation function respectively.

[0084] Among these four path points, the first and fourth points are on the global path, so the angles β of each joint of the collaborative robot in these two positions are i1 , β i4 , angular velocity ω i1 ,ω i4 and angular acceleration α i1 , α i4 The second and third points are the path points selected after replanning by the improved artificial potential field method. The angles of each joint that can be obtained by inverse kinematics are βi2 , β i3 In order to ensure the smooth operation of the collaborative robot arm, it is also necessary to ensure the continuity of the first-order derivatives and second-order derivatives of each joint change. Combining the above constraints, the coefficients of the 3-5-3 polynomial interpolation function can be obtained. The change amplitude of each joint angle, the kinematic and dynamic constraints of each joint of the robot arm, and the maximum movement speed under the safety requirements of human-machine collaboration enable the robot arm to effectively return to the global path while avoiding obstacles and safely reach the target point.

Claims

1. A method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene, characterized in that: The following steps are involved: Step 1: Set the initial pose and target pose according to the collaborative robot’s work task, and perform global path planning in the initial environment; Step 2: The operator completes the assembly task in the assembly area, and the collaborative robot runs according to the predetermined route to complete the assembly task; Step 3: The depth camera obtains the human skeleton and skeleton node information and generates a human body bounding box; Step 4: Determine whether the distance between the arm bounding box and the robot is less than the safe distance. If so, a collision warning is triggered. Step 5: The adaptive obstacle avoidance strategy based on the artificial potential field method replans a new collision-free trajectory, and the robot moves along the new trajectory to avoid obstacles; Step 6: After the obstacle leaves the safety warning range, the robot returns to the global path and finally reaches the target point.

2. According to claim 1, the method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene is characterized by: The global path considers obstacles in the environment as static obstacles in the initial state, and uses the improved RRT algorithm to expand toward the target posture with the initial posture as the root node.

3. According to claim 2, a method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene, characterized in that: The improved RRT algorithm adopts a target bias dynamic sampling strategy, an adaptive step size strategy and path pruning optimization for the RRT algorithm.

4. According to claim 1, the method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene is characterized by: The human skeleton and skeleton nodes are only composed of nodes of the upper body of the human body, and the human body bounding box also only generates a bounding box of the upper body.

5. According to claim 4, the method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene is characterized in that: The collision warning is achieved by simplifying the models of the human body and the collaborative robot arm and adopting the shortest distance collision warning method in combination with the human body motion prediction results.

6. According to claim 5, the method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene is characterized in that: The adaptive obstacle avoidance strategy based on the artificial potential field method proposes an adaptive point selection obstacle avoidance strategy, realizes local obstacle avoidance path replanning, designs a local minimum escape strategy based on plane constraint sampling, and combines piecewise polynomial interpolation to smooth the trajectory of the obstacle avoidance path.

7. According to claim 2, a method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene, characterized in that: The improved RRT algorithm is used to quickly generate a global path, and the cubic spline interpolation function is used in combination with the improved particle swarm algorithm to optimize the trajectory with the shortest time to generate a global trajectory.

8. According to claim 5, a method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene, characterized in that: The camera is used to perceive the human skeleton information in the environment, and the human skeleton time series information is predicted. Combined with hand-eye calibration, the skeleton point information is converted to the robot base coordinate system to perform human-machine shortest distance collision warning detection.

9. The method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene according to claim 6, characterized in that: When the collision warning is not triggered, it runs according to the global trajectory. When the collision warning is triggered, the dangerous part of the original trajectory is eliminated, and the adaptive obstacle avoidance re-planning method is used to generate a new avoidance trajectory. The disconnected trajectories are smoothly connected in combination with the 3-5-3 piecewise polynomial and the new trajectory is input into the robot control system. Finally, the human skeleton information is continuously detected and updated through the intelligent perception system, and the collaborative robot dynamic obstacle avoidance is achieved through collision warning detection and trajectory re-planning, ensuring the safety of man and machine while completing the established work tasks.

10. The method for adaptive obstacle avoidance of a collaborative robot arm in a dynamic scene according to claim 7, characterized in that: The improved particle swarm algorithm dynamically adjusts the inertia weight and learning factor in the particle swarm algorithm to balance the global search and local search, promote the algorithm to explore the global area in the early stage and accurately search the local area in the later stage, ensure the diversity of the particle population, and thus improve the convergence speed and search efficiency.

Images