Swarm-Intelligence-based Multi-Robot Task Allocation

Multi-Robot Task Allocation (MRTA) in an unknown environments is a core problem in automated systems. To complete all tasks in the environment, three distinct steps are needed to be performed: exploration, task allocation, and task execution. We propose a hierarchical algorithm based on swarm intelligence to perform these three steps concurrently in order to improve the overall system performance. Œe proposed algorithm is a heuristic solution for a class of MRTA problems, named ST-MR-IA, which is NP-hard [3]. Background and Contribution While task allocation problems are challenging in general, the level of challenge is raised even further in unknown or dynamic environments. Gerkey and Matarić [3] classi€ed MRTA problems along three dimensions: (1) Single-task robots (ST) versus multi-task robots (MT), referred to capabilities of robots for parallel task execution. (2) Single-robot tasks (SR) versus multi-robot tasks (MR), referred to required number of robots to complete a task. (3) Instantaneous assignment (IA) versus timeextended assignment (TA), referred to the task allocation plan of robots and level of information available for robots. Heuristic approaches recently proposed for MRTA problems, assume either known environments with predetermined number of tasks or single-robot tasks which do not need any cooperation between robots ([3] and references therein). In [2] four di‚erent heuristic approaches were proposed for the task allocation of STMR-TA. In that work, if a robot has completed its share of a task, it communicates the location and the progress of the task to the others, implying that the approach is sequential in nature. We improve the state of the art by addressing the problem of multi-robot tasks in unknown environments: multiple homogeneous robots need to explore an unknown environment and reconnaissance stationary objects (tasks) of interest, as well as to cooperate to locate tasks and execute them. Example Fire €ghting in an unknown environment: tasks are €re sources and should be extinguished by cooperative robots, which do not have a priori knowledge about the environment and tasks’ locations. Œe time required to complete the tasks (€nd, allocate and execute them) is important. Although each task may be completed by one robot, more robots can complete it faster. ProposedmethodWe propose a two-level decentralized algorithm for robots to collaborate to €nd and perform tasks concurrently: (i) At the exploration level of the algorithm, robots use a Bee-Swarm Optimization (BSO) based algorithm to explore the environment and €nd unallocated tasks. Œis algorithm brings randomness to exploration, which is necessary for searching more unknown areas and preventing robots from being trapped in local optimum [1]. When a new unallocated task is found, robots automatically form ∗Œis work was partially supported by the Swedish Foundation for Strategic Research (SSF) through the framework project FiC. †Œis work was supervised by K. Ziarati and M. Eghtesad. ‡I thank M. Papatrianta€lou for comments that greatly improved the manuscript. (a) Iteration 1 (b) Iteration 4000 Figure 1: Two states of the environment with 6 tasks and 10 robots 0 5 10 20 0 0.5 1 1.5 2 ·104 Number of Robots Ite ra tio n hierarchical