未知环境下自主机器人的行为学习研究

STUDY OF BEHAVIOR LEARNING FOR AUTONOMOUS ROBOT UNDER UNKNOWN ENVIRONMENT

机器人为实现在未知环境下的探索任务,必须具有自主学习其行为策略的能力.本文提出了一种自主机器人行为学习机制.机器人通过与环境的交互,基于Q-学习进行行为的自主学习.为降低学习时的计算复杂度,状态空间通过分段映射为不同的类别,从而减少状态—动作对的数量.自主机器人在未知环境中的行为学习是增量式的过程,本文将基于案例的学习与Q-学习结合,使机器人在试错时获得的经验以案例的形式保存,并实现案例库的动态更新.相关案例同时可以降低机器人行为学习时的计算复杂度和试错时的风险.在文中的最后给出了仿真结果.

In order to accomplish exploration task under unknown environment, Robot must have the capability of autonomous behavior learning. In this paper, a mechanism of behavior learning for Robot is proposed. Robot learns its behaviors based on Q-algorithm through interacting with its environment. The state space is segmented into different categories, so that the mumber of state-action pairs is decreased. The behavior learning for Robot under unknown environment is incremental. Case based learning and Q-learning are combined to save the experiences obtained by trial-and-error and to update the case library. Meanwhile, the relevant cases decrease the computational complexity and risk in behavior learning. Finally, the simulation results are presented.