Multi-objective Decentralised Coordination for Teams of Robotic Agents

Abstract

This thesis introduces two novel coordination mechanisms for a team of multiple autonomous decision makers, represented as autonomous robotic agents. Such techniques aim to improve the capabilities of robotic agents, such as unmanned aerial or ground vehicles (UAVs and UGVs), when deployed in real world operations. In particular, the work reported in this thesis focuses on improving the decision making of teams of such robotic agents when deployed in an unknown, and dynamically changing, environment to perform search and rescue operations for lost targets. This problem is well known and studied within both academia and industry and coordination mechanisms for controlling such teams have been studied in both the robotics and the multi-agent systems communities. Within this setting, our first contribution aims at solves a canonical target search problem, in which a team of UAVs is deployed in an environment to search for a lost target. Specifically, we present a novel decentralised coordination approach for teams of UAVs, based on the max-sum algorithm. In more detail, we represent each agent as a UAV, and study the applicability of the max-sum algorithm, a decentralised approximate message passing algorithm, to coordinate a team of multiple UAVs for target search. We benchmark our approach against three state-of-the-art approaches within a simulation environment. The results show that coordination with the max-sum algorithm out-performs a best response algorithm, which represents the state of the art in the coordination of UAVs for search, by up to 26%, an implicitly coordinated approach, where the coordination arises from the agents making decisions based on a common belief, by up to 34% and finally a non-coordinated approach by up to 68%. These results indicate that the max-sum algorithm has the potential to be applied in complex systems operating in dynamic environments. We then move on to tackle coordination in which the team has more than one objective to achieve (e.g. maximise the covered space of the search area, whilst minimising the amount of energy consumed by each UAV). To achieve this shortcoming, we present, as our second contribution, an extension of the max-sum algorithm to compute bounded solutions for problems involving multiple objectives. More precisely, we develop the bounded multi-objective max-sum algorithm (B-MOMS), a novel decentralised coordination algorithm able to solve problems involving multiple objectives while providing guarantees on the solution it recovers. B-MOMS extends the standard max-sum algorithm to compute bounded approximate solutions to multi-objective decentralised constraint optimisation problems (MO-DCOPs). Moreover, we prove the optimality of B-MOMS in acyclic constraint graphs, and derive problem dependent bounds on its approximation ratio when these graphs contain cycles. Finally, we empirically evaluate its performance on a multi-objective extension of the canonical graph colouring problem. In so doing, we demonstrate that, for the settings we consider, the approximation ratio never exceeds $2$, and is typically less than $1.5$ for less-constrained graphs. Moreover, the runtime required by B-MOMS on the problem instances we considered never exceeds $30$ minutes, even for maximally constrained graphs with one hundred agents.

More information

Identifiers

Catalogue record

Export record