Switching Decision of Air-Ground Amphibious Robot using Neural Network-based Reinforcement Learning

Abstract

This paper studies the problem of autonomous decision-making motion mode switching based on external environmental information when the Air-Ground Amphibious Robot is in complex environment. In the process of robot autonomous decision-making, the limitation of motion time and energy consumption should be considered, and the motion state with short motion time and low energy consumption should be selected. In this paper, the methods of reinforcement learning based on neural network are put forward to solve the problem of intelligent switching decision of the Air-Ground Amphibious Robot, aiming at making the robot choose the appropriate mode to move in a certain state of environment. Reinforcement learning based on neural network can not only generate decision function in the process of learning, but also solve the reinforcement problem of continuous environment state space. When a robot performs a task, it is important to reduce the energy consumption of motion. Therefore, this paper takes the energy consumption and motion time as the basis of robot decision-making and the standard of motion evaluation. In this paper, we provide the simulation results and demonstrate the feasibility of our method, which can effectively realize the high-efficiency motion ability of the Air-Ground Amphibious Robot in the complex environment.