Research On Intelligent Planning Method Of Soft Robot Based On End-to-end Reinforcement Learning Model

Because soft robots have a high level of flexibility and motion capabilities,they can perform better than rigid-body robots in complex,unobservable and unstructured environments,so the research community has conducted extensive research on them.And soft materials have very complex nonlinear kinematics properties,there are some challenges in planning and control.In order to make the soft continuum robot intelligently complete the desired tasks.Firstly,we combined the bionics idea in this project to build a flexible robot G-Bot with flexible structure.Secondly,in order to achieve the purpose of intelligent planning and control,In this project,we first proposed an end-to-end model for planning and control of flexible robots.This method is based on a large amount of offline data and uses deep reinforcement learning to plan the optimal trajectory,and finally passes the path information to the actual continuous soft robot Planning control in the arm.Finally,in the simulation and real experimental verification of this subject,it proved that G-Bot has certain maneuverability and intelligently flexibility.At the same time,we propose an end-to-end reinforcement learning model.Optimal path planning based on RGB images can achieve simulation and practical verification success.The model utilizes a large amount of off-line experienced visual data and deep neural networks to process the best path,and complete the task without performing difficult and precise kinematic analysis.The verification experiment results show that the end-to-end model can achieve excellent performance in specific tasks.The results also prove that deep reinforcement learning has great application potential in solving the planning and control problems of nonlinear software robots.