自组织映射神经网络量化机器人强化学习方法研究

Research on Reinforcement Learning of the Robot Based on The SOM Neural Nebnwork Quantization

强化学习一词来自于行为心理学 ,这门学科把行为学习看成反复试验的过程 ,从而把环境状态映射成相应的动作 .在设计智能机器人过程中 ,如何来实现行为主义的思想、在与环境的交互中学习行为动作 ?文中把机器人在未知环境中为躲避障碍所采取的动作看作一种行为 ,采用强化学习方法来实现智能机器人避碰行为学习 .为了提高机器人学习速度 ,在机器人局部路径规划中的状态空间量化就显得十分重要 .本文采用自组织映射网络的方法来进行空间的量化 .由于自组织映射网络本身所具有的自组织特性 ,使得它在进行空间量化时就能够较好地解决适应性灵活性问题 ,本文在对状态空间进行自组织量化的基础方法上 ,采用强化学习 .解决了机器人避碰行为的学习问题 。

The occept of the reinforcement learning comes from behavior psychology that takes behavior learning as trial and error ,by which the states of envirment are mapped into corresponding actiovs .There's a question of how dose the beheviorism be used to learn the actions in interactoion with the environment in destgning intelligent robot in this paper ,the actions that robot takes to avoid obstades are taken as one class of behaviors and the reinforcement learning is used to realize behavior learning of obstacle avoidance.The quantization of state space shows very important in improving robot's learning speed .The SOM neural network is adopted to get quantization of state space in this paper ,The self organization characteristic of SOM neural network makes it possible to solve adaptation and flexible in space quantization .The reinforcement learning is used to solve the robot learning of avoidence collision behavior based on quantization of state space and the satisfying results are got .