Evolution of Kiting Behavior in a Two Player Combat Problem

摘要

We examine the use of an evolutionary algorithm to design a feedback controller for a two player combat problem. This problem consists of two players on a one-dimensional line. One player is referred to as the Defender and is held at the origin (unable to move). The other player is referred to as the Attacker and is free to move back and forth with a constant speed. The goal of the Attacker is to inflict as much damage as it can on the Defender while suffering as little damage as possible. The greater the difference between damage inflicted vs damage taken, the more successful the attack. The Attacker’s controller is represented by a parameterized control vector. An evolutionary algorithm is then used to evolve populations of these control vectors in an attempt to obtain a near optimal feedback controller.