Research On System Design And Control Strategy Of Self-balancing Exoskeleton Robot

Currently,the most popular exoskeletons are underdriven exoskeletons,which provide active assistance for single or multi-joint movements in the sagittal plane of the body,but they do not allow for self-balancing walking assistance.The wearer with a motion disability must rely on external support such as crutches or a trolley to maintain balance.This underdriven exoskeleton can only be used for paraplegics with ablebodied upper limbs and cannot provide rehabilitation assistance for the much larger number of quadriplegics.In addition,due to the lack of active degrees of freedom,this type of exoskeleton robot is also unable to provide a gait that is highly compatible with the human body,which can lead to problems such as poor wearing comfort and unnatural walking.The fully actuated self-balancing exoskeleton is a new type of exoskeleton that not only provides self-balancing assistance to paraplegic,hemiplegic,and tetraplegic patients,but also solves the problems of unnatural gait and poor wearing comfort.However,little research has been conducted on this type of exoskeleton,and the existing research suffers from the following problems: insufficient coupling between human and exoskeleton due to axis deviation;lack of highly safe,natural,and anthropomorphic path and gait planning strategies;and large-scale motion deviation due to the complexity of the system structure and wearing specificity.To address these shortcomings,this thesis aims to develop a fully driven,self-balancing exoskeleton robot with high human-robot compatibility,natural gait output,greater intelligence,and ease of use.The focus is on three scientific problems: "A multi-joint motion bionic approach with low axis deviation","A model-free motion compensation strategy for complex structured robots",and "An end-to-end motion planning control strategy for human-like thinking".This paper conducts research in four main areas: "synthesis and structure creation for highly coupled human-machine mechanisms","expressions for kinematic models and motion constraint spaces in different gait phases","motion error compensation strategies based on meta-heuristic optimization algorithms combined with neural networks",and "multi-level motion planning control strategies based on deep reinforcement learning combined with model-based algorithms".The following findings have been obtained:(1)In terms of exoskeleton system design,the self-balancing exoskeleton is defined as a synthesis of a wearable exoskeleton and a bipedal robot.The functional requirements of a tetraplegic patient were analyzed and determined,and then the overall system framework of the self-balancing exoskeleton robot was designed.By analyzing the anatomy of human movement and statistical data,the number of active degrees of freedom,joint layout,range of motion and dimensional parameters required for the exoskeleton are determined.Through mechanism synthesis and structure creation,the problems of multi-joint motion coupling and human-machine axis deviation are solved.A 10 degrees of freedom self-balancing exoskeleton robot with a serial mechanism,Auto LEE-G1,and a 12 degrees of freedom self-balancing exoskeleton robot with a serial-parallel hybrid mechanism,Auto LEE-G2,have been designed.(2)In the analysis of exoskeleton systems,simplified kinematic models of Auto LEE-G1 and Auto LEE-G2 are established and the similarities and differences between them are analyzed by geometric constraints.The analytical solutions of the forward and inverse kinematic models of the exoskeleton are obtained using the chain rule and geometric method respectively,providing a model basis for fast,real-time and accurate exoskeleton motion planning.To ensure the safety of the exoskeleton movement,the kinematic constraint characteristics of the exoskeleton are analyzed in terms of two gait phases,one for the single-leg support phase and the other for the double-leg support phase.The expressions of the exoskeleton movement constraint space are obtained using the geometric constraint relations.(3)In terms of exoskeleton motion error compensation control,Auto LEE-G2 is used as the object of study.Based on the D-H kinematic modelling method,the geometric and non-geometric error source parameters of the exoskeleton are represented by the 48-dimensional constant matrix and the variable matrix respectively,and the analysis shows that the non-geometric error is the main error source of the exoskeleton.An extreme learning machine network model optimized by an modified mayfly algorithm is proposed to regress the mapping relationship between the exoskeleton motion error and the input joint angles.A motion error compensation strategy is designed based on this mapping model,and finally a comparison experiment is used to verify that this compensation strategy can effectively reduce the motion error.(4)In terms of exoskeleton motion planning control,a multi-level end-to-end fully autonomous trajectory planning algorithm framework is proposed,which can realize the conversion of rough high-level motion intention into concrete joint trajectories.The footholds path planning layer combines motion constraint space and artificial potential field theory into a deep reinforcement learning framework,which can output safe,human-like discrete footholds sequences online.The gait trajectory planning layer utilizes ZMP and model predictive control theories to rapidly generate self-balancing gait trajectories.The joint trajectory planning layer uses model predictive theory and neural network to design the ZMP compensator and kinetic filter to obtain gait trajectories with higher stability margins,which can be converted into joint trajectory curves to drive the exoskeleton robot.