Flexible Adaptive Control of Snake-Like Robot Based on LSTM and Gait

Abstract

This paper presents a flexible control method for snake-like robots to adapt to the environment autonomously. This method enables the snake-like robot to sense the environment and adjust autonomously according to the changes of the environment. For example, snake-like robots climb pipes with varying diameters. In order to achieve efficient and flexible motion, we adopt closed-loop gait control. This control method uses a parameterized sine wave (gait function) and long short-term memory network (LSTM) model. Because of the structure of LSTM is suitable for prediction of time series data, we use LSTM to predict the changes of joint angles that can best represent the shape change of snake-like robot, and integrate the predicted joint angle values with the gait values to realize the control of robot. Because this control method will eventually allow the snake-like robot to move autonomously in the changing environment, we can achieve the flexible adaptive behavior of the robot.