A Market-Based Framework for Tightly-Coupled Planned Coordination in Multirobot Teams

This dissertation explores the challenges of one of the most difficult classes of real-world tasks for multirobot teams: those that require long-term planning of tightly-coordinated actions between teammates. These tasks involve solving a distributed multi-agent planning problem in which the actions of robots are tightly coupled. Moreover, because of uncertainty in the environment and the team, robots must frequently replan and closely coordinate with each other throughout execution. We have developed a coordination framework called Hoplites in response to the need for effective approaches to these problems. Although planning for tightly-coupled multirobot systems is a difficult problem, Hoplites solves this problem efficiently by using distributed decision-making whenever possible and centralized planning as required. Hoplites is a market-based system that consists of passive coordination and active coordination mechanisms which are tailored to easier and harder problem scenarios, respectively. Passive coordination is light on computation and communication and allows teammates to iteratively respond to each other’s actions without directly influencing them. When passive coordination traps robots in local minima, active coordination improves solutions by enabling robots to influence each other directly by buying each other’s participation in complex plans over the market. Because Hoplites selectively injects pockets of complex coordination into the system, it provides these improvements while remaining computationally competitive with other distributed approaches. Moreover, this selective complexity allows Hoplites to outperform centralized approaches as well because it can often exploit planners with performance guarantees which a centralized approach cannot. Additionally, Hoplites is widely applicable to real-world problems because it is general, computationally feasible, scalable, operates under uncertainty, and improves solutions with new information. This dissertation makes a number of contributions to the literature. First, it develops Hoplites, a general and adaptive approach to these complex problems. Second, it formalizes the problem space which, in turn, enables us to describe and share solutions between domains that may have previously appeared disparate. Third, it is the first direct application of market-based approaches to tight coordination. Fourth, it presents the first evaluation of and recommendations for planning algorithms for tight coordination. Lastly, it improves the previous state-of-the-art coordination framework for these problems.