摘要
递阶强化学习是解决状态空间庞大的复杂系统智能体决策的有效方法。通过引入启发式算法思想,对一种递阶强化学习方法进行改进,使得智能体在学习过程中融入了历史信息,提高了学习效率,解决了在庞大状态空间和动态变化环境中对智能体进行最优行为策略学习的问题。以扩展的信念、愿望和意图意识模型为基础,提出了一种具有主动性、自治性、反应性、社会性的自主机器人路径规划智能体体系结构,通过仿真实验,证明了路径规划智能体的可行性和有效性。
Hierarchical reinforcement learning was an effective method to solve decision problems for complex system agent with enormous number of states. By introducing heuristic algorithm, a hierarchical reinforcement learning method was improved, making the agent obtain historical information in the learning process to increase the learning efficiency so as to solve the optimal strategy of agent learning problem in large-scale state space and dynamic environment. Based on expanded Belief Desire Intention (BDI) model, the architecture of autonomous robot path planning agent was presented with properties as initiative, autonomy, reactivity and sociability. It was proved by simulation that the path planning agent was feasible and effective.
出处
《计算机集成制造系统》
EI
CSCD
北大核心
2009年第6期1215-1221,共7页
Computer Integrated Manufacturing Systems
基金
国家973计划资助项目(2007CB714701)~~
关键词
智能体
强化学习
意识模型
路径规划
agent
reinforcement learning
consciousness model
path planning