摘要
通过引入一种启发式学习算法,部分改进了MAXQ递阶强化学习方法,并结合模糊神经网络开发了一种自主式地面车辆(ALV)全局路径规划Agent。该智能Agent充分融合了人类操作经验和机器学习能力,为强化学习明确了搜索方向,缩减了计算量,具有较强的自适应能力,满足了系统的实时性要求。仿真结果表明:在庞大状态空间和动态变化环境中,全局路径规划Agent能够有效、实时地进行最优行为的策略学习。
By introducing FMQ（frequency maximum Q） heuristic learning algorithm, a hierarchical method of reinforcement learning was improved, through the combination of this method and fuzzy neural--networks, a global path planning Agent was developed. This Agent integrated the human op-eration experience and the capacity of machine learning, so that it ensured the search direction, reduced the amount of computation, strengthened the ments. The simulation results show that the global effective and real--time in the large state space and adaptive capacity and met the real--time requirepath planning Agent can find the optimal strategy the dynamic changing environment.
出处
《中国机械工程》
EI
CAS
CSCD
北大核心
2009年第21期2536-2541,共6页
China Mechanical Engineering
基金
国家重点基础研究发展计划资助项目(2007CB714701)
