基于深度强化学习的移动机器人导航策略研究

Research on Navigation Strategy of Mobile Robot Based on Deep Reinforcement Learning

摘要:

针对移动机器人在复杂动态变化的环境下导航的局限性,采用了一种将深度学习和强化学习结合起来的深度强化学习方法。研究以在OpenCV平台下搭建的仿真环境的图像作为输入数据,输入至TensorFlow创建的卷积神经网络模型中处理,提取其中的机器人的动作状态信息,结合强化学习的决策能力求出最佳导航策略。仿真实验结果表明：在经过深度强化学习的方法训练后,移动机器人在环境发生了部分场景变化时,依然能够实现随机起点到随机终点的高效准确的导航。

Abstract:

Aiming at the limitation of mobile robot navigation in complex and dynamic environment, a deep reinforcement learning method combining deep learning and reinforcement learning is adopted. In this study, the image of the simulation environment built under the OpenCV platform was used as input data, which was input into the convolutional neural network model created by TensorFlow for processing, in which the robot"s action state information was extracted, and the optimal navigation strategy was obtained by combining the decision-making ability of reinforcement learning. The simulation results show that after the training with the method of deep reinforcement learning, the mobile robot can still realize the efficient and accurate navigation from the random starting point to the random ending point when some scenes change in the environment.