High-speed gait generation method for underwater snake-like robot based on proximal strategy optimization

Abstract

A high-speed gait generation method for an underwater snake-like robot based on proximal strategy optimization, comprising: building a training environment for reinforcement learning, including a reset() function and a step() function; implementing a proximal strategy in the training environment for reinforcement learning Optimization algorithm; use the sine function as the basis function to fit the weight obtained by the policy network, and then convert it into the gait parameters of the underwater snake robot. The beneficial effects of the present invention are as follows: the gait generation method of the present invention takes high speed as the optimization target, and when the gait motion obtained by optimizing the underwater snake robot under the gait generation method, its forward motion speed is higher than that of the traditional gait method The forward movement speed of the optimized optimal gait movement is faster; the gait generation method of the present invention will search in a larger range than the serpentine gait equation during the optimization process, and the obtained gait equation form is often It will be different from the serpentine gait equation, which is a new type of locomotion gait.

Description

technical field [0001] The invention relates to a method for generating high-speed gait of a robot. In particular, it relates to a high-speed gait generation method for an underwater snake robot based on proximal policy optimization. Background technique [0002] Due to the complex and changeable underwater environment, traditional underwater detection robots are usually difficult to deal with, and the problem of short battery life is common. In recent years, robots designed based on bionics have performed very well in the problem of adapting to complex environments, and their movement methods are derived from biological mechanisms, so the energy efficiency ratio is much higher than that of traditional robots. The design concept of the underwater snake-shaped robot studied in the present invention is derived from biological snakes. Like biological snakes, it has redundant joint structure, extremely high flexibility, and can move in various environments with high movement ef...

AI Technical Summary

Problems solved by technology

Aiming at the lack of motion control of the traditional gait of land snake robots in an unstructured environment, researchers at Carnegie Mellon University adopted an asynchronous advantage actor-critic (asynchronous advantage actor-critic, A3C) reinforcement learning algorithm under a distributed framework, The movement of the snake robot on the unstructured terrain is realized online and offline respectively, and the movement efficiency is improved by 40%, but the method still uses the prior knowledge of the traditional gait equation

Images