Multi-agent coordination techniques for multi-robot task allocation and multi-robot area coverage

Multi-robot systems have emerged as a central research theme within robotics with applications in several domains that require automated robotic assistants such as unmanned search and rescue, automated surveillance and reconnaissance operations, automated civilian transportation, extra-terrestrial exploration, and even domestic applications such as agriculture, lawn mowing, vacuum cleaning, etc. One of the major computational challenges in multi-robot systems is to design appropriate coordination techniques between the robots that enable them to perform operations efficiently in terms of time, cost and energy expended while keeping the system robust to individual robot failures as well as scalable in the number of robots. Coordination technologies from the field of multi-agent systems offer a rich array of solutions that can be adapted to multi-robot sytems. In this talk, we will summarize our research on two operations that are frequently encountered in many multi-robot domains, namely, multi-robot area coverage [1] and multi-robot task allocation [2]. First, we will describe a coalition game theory based technique for dynamically reconfiguring multi-robot teams by splitting or merging them, when they encounter obstacles while covering an initially unknown environment. We will introduce two heuristics that, given the set of robots requiring reconfiguration, guarantee rapid convergence to the appropriate partition of the set of robots while taking into consideration the physical characteristics of the robots and the environment [3]. Secondly, I will describe a distributed, auction-based multirobot task allocation algorithm called DynamicBids [4] that improves the performance of tasks and significantly reduces the communication overhead between robots by allowing a bidder robot to selectively revise its bids on tasks if that improves the cost of the schedule of the tasks to the bidder robot. For both techniques, we will describe analytical results, and, experimental results from simulations on the Webots simulator as well as on physical robots. Finally, we will demonstrate our ongoing work on autonomous, multi-robot landmine detection that uses the techniques mentioned above to coordinate a set of robots, fitted with different types of landmine detection sensors, to potentially improve the accuracy with which landmines can be detected [5].