基于深度自动编码器与Q学习的移动机器人路径规划方法

Mobile Robot Path Planning Based on Deep Auto-encoder and Q-learning

针对移动机器人在静态未知环境中的路径规划问题,提出了一种将深度自动编码器(deep auto-encoder)与Q学习算法相结合的路径规划方法,即DAE-Q路径规划方法.利用深度自动编码器处理原始图像数据可得到移动机器人所处环境的特征信息;Q学习算法根据环境信息选择机器人要执行的动作,机器人移动到新的位置,改变其所处环境.机器人通过与环境的交互,实现自主学习.深度自动编码器与Q学习算法相结合,使系统可以处理原始图像数据并自主提取图像特征,提高了系统的自主性;同时,采用改进后的Q学习算法提高了系统收敛速度,缩短了学习时间.仿真实验验证了此方法的有效性.

To solve the path planning problem of mobile robot in static unknown environment, a new pathplanning method was proposed which combined the deep autoencoder with the Qlearning algorithm,namely the DAEQ path planning method. The deep autoencoder processed the raw image data to get thefeature information of the environment. The Qlearning algorithm chose an action according to theenvironmental information and the robot moved to a new position, changing the surrounding environmentof the mobile robot. The robot realized autonomous learning through the interaction with the environment.The system processed raw image data and extracted the image feature autonomously by combining thedeep autoencoder and the Qlearning algorithm, and the autonomy of the system was improved. Inaddition, an improved Qlearning algorithm to improve the system爷s convergence speed and shorten thelearning time was utilized. Experimental evaluation validates the effectiveness of the method.