FPGA implementation of multiple Pursuit-Evasion games with decentralized Learning Automata

Abstract

Listen

Translate article

This paper addresses the implementation of multiple Pursuit-Evasion (PE) games using Field Programmable Gate Array (FPGA) technology. The multi-agent game is modeled as Markov chains with each player working as a decentralized unit and using Learning Automata (LA). To take a desired action at each step for each player, an efficient Learning algorithm is used that leads to the players to evolve and adapt to the environment in order to solve difficult problems. To realize the PE game in the hardware devices, such as FPGAs in this paper, the system is optimized and designed based on the properties of the hardware technology. The implementation approaches for the realization of the main building blocks of the system are presented in detail. A modified Learning algorithm is used in the hardware implementation. This system has been developed in VHSIC Hardware Description Language (VHDL) and implemented using Xilinx Virtex 6 FPGAs. The simulation results have been achieved and presented in this paper. To prove the efficiency of the Learning algorithm designed with hardware technology, the simulation results are also presented in statistic version, which further proves that the speed of capture is decreased after using the Learning algorithm and finally converges to an equilibrium point in this multiple PE games.