Intelligent pursuit & evasion in an unknown environment

Abstract

This paper introduces a novel and flexible simulation platform for studying pursuit and evasion in unknown 2-D environments of arbitrary obstacles, in an effort to expand the practical application of pursuit-evasion research. The platform provides realistic simulation of the sensing capability of each robotic agent (either a pursuer or an evader). Each agent uses real-time local sensing to collect information from the environment while it simultaneously plans and executes its motion to best satisfy one or more objectives. The evader's objectives are to reach a specific goal location as quickly as possible and to avoid being caught by the pursuer. The pursuer's objectives are to locate and capture the evader whose motion is unknown, and when the evader is not seen, explore the environment and predict where the evader may be. Under a common real-time planning paradigm, each agent's planner dynamically adapts its goals and objectives to the agent's changing circumstances so that the agent can always choose the best course of action. Simulation results have shown that the introduced approach is an effective means to study sophisticated pursuit-evasion scenarios and accomplish objectives for both the pursuer and the evader in an unknown environment. The platform can be easily expanded to accommodate multiple agents in more complex pursuit-evasion tasks.