Target Detection and Interception Using Modified POMDP Based Motion Planning of Mobile Robots in a Known Environment

Abstract

This thesis presents an approach to locate an adversarial, mobile evader in an indoor environment using motion planning of mobile pursuers. The approach presented in this thesis uses motion planning of mobile robots to search a target in a graph and clear the workspace. The algorithm used is Modified Partially Observable Markov Decision Process (POMDP), a probabilistic search method to clear the indoor workspace in a pursuit evasion domain. In this thesis, the indoor environment is assumed to be known beforehand and the mobile evader to be adversarial with no motion model given. The workspace is first discretized and then converted to a graph, whose nodes represent the rooms and corridors and edges represent connection between them. The task of pursuer is to clear the whole graph with no contaminated node left in minimum possible steps. Such path planning problems are NP-hard and the problem scales exponentially with increased number of pursuers and complex graph.