Motion Planning For Manipulator With Pushable Obstacles

Motion planning for manipulator is a fundamental problem in robotics,which seek a trajectory from the initial state to the target state in safety.In the environments with obstacles,it’s usually to find a collision-free path.However,in cluttered environments where there is no collision-free path,the obstacles must be rearranged.An effective way to rearrange the obstacles is to push the obstacles by any part of the manipulator.This paper presents a novel motion planning method called motion planning for manipulator with pushable obstacle that enables the manipulator to push the obstacles by any part of the manipulator in safety during the motion towards to the target.The advantages of this algorithm are: the distribution of free space is changed by pushing the obstacle,which can solve several problems that can not be solved by the collision-free motion planning;the manipulator moves towards to the target and pushes obstacles away incidentally,and the efficiency is improved.This thesis investigates the modeling of pushing interaction and the design of motion planning algorithm of motion planning for manipulator with pushable obstacle,mainly including the following three aspects:(i)The definition of motion planning algorithm of motion planning for manipulator with pushable obstacle is introduced,the pipeline of the approach of the motion planning is discussed,and the challenge of motion planning algorithm of motion planning for manipulator with pushable obstacle is analyzed.Two model of pushing interaction is present:the risk distribution model,which describes the relation between the position of the point where the manipulator contacts the obstacle,and the risk of damaging the obstacle or the manipulator;the pushing motion model,by which the trajectory of obstacle can be predicted with the trajectory of manipulator and the initial pose of the obstacle.(ii)The motion planning algorithm based on risk distribution is present.The planning method,which integrates the sampling-based motion planning algorithm and flexible obstacle avoidance based on risk distribution,is designed to plan a trajectory which passes through the low-risk areas to push the obstacles and avoids the high-risk areas to guaran-tee the safety.This method can solve the motion planning problem using flexible obstacle avoidance.Finally,the effectiveness of the proposed method is verified by simulations and real robot experiments.(iii)A model learning method for motion planning algorithm of motion planning for manipulator with pushable obstacle is proposed.In the process of manually modeling the risk distribution models of obstacles,there are several assumption and simplification,which leads to the errors in the model.In order to solve this problem,we propose a learningbased method to establish the models of the motions of pushed obstacles,so that the models are trained by repeated pushing the obstacle to avoid the error of manually modeling.The model learning approach is proposed to train the pushing motion model of obstacle from experiments,and use this model to predict the motion of obstacle interacting with manipulator in motion planning.We demonstrate our results with experiments in simulation and real robot.The experiment results indicate that comparing with motion planning with traditional obstacle avoidance,our method solves and executes successfully in the environment without a collision-free path.