A Task Decomposition Using (HDec-POSMDPs) Approach for Multi-robot Exploration and Fire Searching

Abstract

In this paper, hierarchical control architecture for coordinated multi-robot systems (MRS) task decomposition is presented; based on a hybrid decentralized Partially Observable Semi-Markov Decision Processes (HDec-POSMDPs).  In this architecture, robots can make their own decisions according to their locally collected data with limited communication between a robot team.  In this proposed architecture, the global task is decomposed into multiple local sub-tasks using divide and conquer design, each task is described as a set of regular languages. MRS are modeled as a discrete event system and each robot is represented by a deterministic finite state automaton model. Direct Cross-Entropy (DICE) can be used for searching the space of the best frontier cells to solve the Dec-POSMDP and each sub-task is assigned to one or more robots to be executed. The proposed algorithm is implemented, tested and evaluated in the computer simulator. By using this architecture, the task execution time is minimized, the fire sources cluttered in an environment have been searched in an effective manner and the performance of MRS has been enhanced with respect to energy consumption and communication load; when they are used for exploring different environments as well as when they are used for detecting the sources of the fire and reporting about them.