Research On The Motion Planning Method Of The Manipulator For Moving Object Capture Task

Compared with static object capture,the robot arm’s ability to capture the moving object can significantly improve work efficiency and manmachine or multi-machine cooperation performance and has a wide range of application prospects.The two main techniques for moving object capture are visual servo control and motion planning.The local action generated by the servo controller will fall into the extreme value,and the controller will not generate subsequent action when the target is blocked.The motion planning method can reduce the risk of the manipulator falling into the local extreme value and has stronger robustness to the occlusive target.However,the calculation speed of the current motion planning method cannot cope with the changeable environment.Therefore,aiming at the capture task of the moving target object with a linear velocity of less than 0.15m/s in complex and changeable scenes,this article conducts an in-depth study on the motion planning method of the 6-DOF manipulator.The main research contents and innovative achievements are as follows:Aiming at the problem that the existing path planning algorithm needs to set a fixed grasping position when capturing the moving object,which leads to reduced integrity,this article proposes an improved sampling path planning algorithm L-Connect under time constraints.The algorithm can predict the intersection point of the moving object multiple times,only sample in the collision area,and directly generate waypoints in the space without collision to improve the success rate and computational efficiency.Aiming at the shortcoming of the existing manipulator trajectory optimization algorithm that takes a long time to calculate and is unsuitable for capturing the moving object,this article proposes the EGO-ARM trajectory optimization algorithm.The B-spline curve represents the joint trajectory,and the convex hull characteristics of the B-spline curve nonlinearly optimize the curve’s control points.The EGO-ARM algorithm introduces the EGO algorithm in the workspace into the joint space of the manipulator.It only builds a local approximate gradient field in the joint space to significantly improve the efficiency of trajectory optimization.Based on ROS simulation and prototype experiments,the algorithms proposed in this article can quickly generate smooth trajectories in the moving object capture task in environments with fixed obstacles and the moving obstacle following the target.These experiments verify the high efficiency and robustness of the two algorithms.This article proposes a new multi-module sequential sense-plan-act algorithm to solve the problem of rapid online trajectory replanning due to the continuous change of the environment.The motion prediction module in the new algorithm can predict the motion trajectory by fitting the historical path of the moving object,and it uses the predicted trajectory of the object for trajectory planning,which increases the effective length of the planned trajectory.The grasp planning module in the new algorithm uses a clustering algorithm to combine approximate grasp poses while ensuring the diversity of grasp poses to improve the operation efficiency.The trajectory planning module generates trajectories for multiple grasp poses and executes the time-optimal trajectory to shorten the moving object capture time compared with the algorithm that only uses one optimal grasp pose for trajectory planning.The trajectory planning module streamlines the previously proposed motion planning algorithm and can generate a joint trajectory satisfying the constraints within an average of 2ms.When the object’s trajectory cannot be predicted for long,the new algorithm can plan real-time trajectory based on short-term motion prediction information.This article designs various test conditions based on ROS to verify the effectiveness of each module of the new algorithm.Prototype experiments of the conveyor belt moving target capture,target capture in random environments,and space object capture are designed,respectively,and the accuracy and feasibility of the new algorithm are verified.In order to solve the problem that the middle and back parts of many global trajectories fail before execution due to unstable sensor information or abrupt environmental changes,this article proposes a local smooth trajectory generation algorithm for the manipulator based on deep imitation learning.Firstly,the B-spline curve control point sequence generated by the new sequential sense-plan-act algorithm and the depth images at the corresponding time together forms the expert data set.Then,we establish a Transformer-based policy network and design a loss function including trajectory smoothness,dynamic feasible,collision,and trajectory similarity term.Finally,the policy network is trained using the expert dataset.The image sequence is input into the policy network,and the control point sequences of multiple local trajectories can be output.The self-attention mechanism in the policy network establishes the dependence between the sequence of sequential output control points,and a small number of control points can represent a long,smooth,and dynamically feasible trajectory.Through the Pybullet-based simulation experiment and prototype experiment,the feasibility of the algorithm in the conveyor belt moving target capture task is verified,which lays the foundation for developing more efficient and intelligent motion planning algorithms.