Reward function establishing method based on walking ratio trend change

Abstract

The invention discloses a method for establishing a reward function based on walking ratio trend change. The method comprises the following steps: calculating the step length D of a wearer of an exoskeleton robot; calculating a gait period T(k); calculating a walking ratio W according to the step length D and the gait period T(k); establishing a walk ratio sampling sequence and scoring the sampling sequence in the walk ratio sampling sequence; establishing a reward function model. The reward function model based on walking ratio trend change can be applied to an algorithm for optimizing exoskeleton parameters, the efficiency of reinforcement learning is enhanced, and rapid convergence of the exoskeleton parameters is promoted.

Description

[0001] (1) Technical field: [0002] The invention belongs to the technical field related to robots, and is a method for establishing a walking ratio reward function of a gait rehabilitation flexible exoskeleton robot, which can be applied to the adaptive control task of control parameters of a flexible exoskeleton based on a reinforcement learning method. [0003] (two) background technology: [0004] The flexible exoskeleton robot can help the elderly with inconvenient legs and feet to walk and enhance the strength of human legs. It has a wide range of uses in rehabilitation treatment, daily travel, etc. Due to the large individual differences between people, at present, most of the control parameters of exoskeleton robots need to be adjusted according to the wearer's own motion characteristics, which is time-consuming and labor-intensive and cannot track the wearer's body changes. [0005] Reinforcement learning can find the optimal strategy in the interaction with the envi...

AI Technical Summary

Problems solved by technology

However, imitation learning requires alternating iterations of Inverse Reinforcement Learning and reinforcement learning. The process is too complicated, and imitation learning relies on expert samples, which is not applicable to some occasions where expert samples are lacking.

In addition, for some problems with sparse rewards, the efficiency of reinforcement learning is also very low. Researchers have proposed some solutions for this, including setting auxiliary tasks and introducing curiosity mechanisms. Tasks are provided by experts with corresponding prior information, which cannot solve the sparse reward problem of reinforcement learning in a general sense

Images