基于多组并行深度Q网络的连续空间追逃博弈算法

为解决连续空间追逃博弈(PEG)问题,提出一种基于多组并行深度Q网络(DQN)的连续空间PEG算法.应对连续行为空间中为避免传统强化学习存在的维数灾难不足,通过构建Tak-agi-Sugeno-Kang模糊推理模型来表征连续空间;为应对离散动作集自学习复杂且耗时不足,设计基于多组并行DQN的PEG算法.以4轮战车PEG...

Continuous Space Pursuit-evasion Game Algorithm Based on Multi-group Deep Q Network

A continuous space pursuit-evasion game algorithm based on multi-group deep reinforcement learning is proposed to solve the problems in continuous space pursuit-evasion game(PEG). In order to avoid the insufficient curse of dimensionality of continuous space in traditional reinforcement learning，a TSK fuzzy inference model is established to represent the continuous space.And a pursuit-evasion game algorithm based on multi-group deep reinforcement learning is designed for the complex and time-consuming problems of discrete action self-learning.The simulation environment and motion model were designed by taking the PEG problem of a four-wheel vehicle as an example, and the simulation experiments were carried out with Q-learning algorithm, reinforcement learning algorithm based on qualification trace and genetic algorithm based on reward, respectively. The simulated results show that the continuous space PEG algorithm can be used to solve the problem of continuous space pursuit-evasion game well，and continuously improve the ability to address problems with the increase in learning times，and has the comparative advantages of less time consuming for independent learning and short application time.