Imitation Learning Based Motion Planning And Trajectory Tracking Control For Collaborative Robots

Cobots are robots that can interact directly and closely with humans in shared spaces.Due to their safety,low cost,flexibility in deployment,and simplicity of operation,they have been widely used in various fields such as industrial manufacturing,medical rehabilitation,warehouse logistics,and home services in recent years.However,as cobots move from simple environments and lowdifficulty tasks to complex non-structured environments and high-difficulty tasks,coupled with their lower rigidity resulting in poor repeatability of positioning accuracy,they currently face the challenges of difficult motion planning and highaccuracy trajectory tracking control.Therefore,this paper focuses on the research of cobot imitation learning motion planning technology and high-accuracy trajectory tracking control technology for flexible-joint cobots.Firstly,this paper proposes two improved imitation learning motion planning algorithms.The topological equivalence of the Dynamic Movement Primitives(DMP)algorithm is introduced into the Kernelized Movement Primitives(KMP)algorithm,and the DS-KMP algorithm is proposed to combine the KMP algorithm with the DMP algorithm.The DS-KMP algorithm solves the problem of difficult overall trajectory modulation of the KMP algorithm.Additionally,this paper proposes an improved Auxiliary Classifier Generative Adversarial Imitation Learning(ACGAIL)algorithm to solve the problems of mode collapse,gradient vanishing,and low efficiency of expert trajectory utilization found in the generative adversarial imitation learning algorithm.Secondly,this paper proposes an adaptive neural network trajectory tracking controller for high-rigidity flexible-joint cobots(ANN-F controller),which realizes high-accuracy tracking of the results of imitation learning motion planning.Further,the proposed algorithms are applied to practical cobot systems.The proposed DS-KMP algorithm is used to adaptively learn object receiving skills based on online prediction of human giver motion trajectories,meet safety constraints and ensure human givers to perform handover operations naturally.The proposed ANN-F controller is used for high-accuracy real-time tracking of the results of imitation learning motion planning.Then,efficient and natural human-tocobot dual-arm handover can be achieved.The improved ACGAIL algorithm is used to achieve intelligent chemical synthesis by cobots.Through experimental verification and practical application,the proposed imitation learning motion planning and trajectory tracking control algorithm can effectively solve the problems of motion planning and high-precision trajectory tracking control faced by cobots.