eduzhai > Applied Sciences > Engineering >

Mean Field Game and Decentralized Intelligent Adaptive Pursuit Evasion Strategy for Massive Multi-Agent System under Uncertain Environment

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 29 pages

Abstract: In this paper, a novel decentralized intelligent adaptive optimal strategyhas been developed to solve the pursuit-evasion game for massive Multi-AgentSystems (MAS) under uncertain environment. Existing strategies forpursuit-evasion games are neither efficient nor practical for large populationmulti-agent system due to the notorious "Curse of dimensionality " andcommunication limit while the agent population is large. To overcome thesechallenges, the emerging mean field game theory is adopted and furtherintegrated with reinforcement learning to develop a novel decentralizedintelligent adaptive strategy with a new type of adaptive dynamic programingarchitecture named the Actor-Critic-Mass (ACM). Through online approximatingthe solution of the coupled mean field equations, the developed strategy canobtain the optimal pursuit-evasion policy even for massive MAS under uncertainenvironment. In the proposed ACM learning based strategy, each agent maintainsfive neural networks, which are 1) the critic neural network to approximate thesolution of the HJI equation for each individual agent; 2) the mass neuralnetwork to estimate the population density function (i.e., mass) of the group;3) the actor neural network to approximate the decentralized optimal strategy,and 4) two more neural networks are designed to estimate the opponents groupmass as well as the optimal cost function. Eventually, a comprehensivenumerical simulation has been provided to demonstrate the effectiveness of thedesigned strategy.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×