A Neural Network based Mobile Robot Navigation Approach using Reinforcement Learning Parameter Tuning Mechanism

Abstract

Robotic navigation mechanism is one of the most adhoc research topics on mobile robots, which requires a robot to find a suitable path and move from its current position to a destination position without colliding with any obstacles. This paper employs the intelligent method of reinforcement learning to explore a solution to address the aforementioned problem. It considers the laser beam detected distances and the relative movement angle as the input of neural network model, and the robot’s action posture is denoted as the output. This neural network model is trained by a deep Q-learning network (DQN) algorithm via positive and negative feedback rewards defined by task-specific learning goals. In this sense, the trained model helps the robot determine the appropriate action to take at each state to safely reach the destination point without any manually interference. According to the results on a simulation platform, the trained neural network model makes the robot move from random starting point to random destination successfully, which proves the effectiveness of DQN algorithm in the field of robot navigation.