This presentation delves into techniques that facilitate coordination and cooperation among agents within multi-agent systems. These decision-making agents interact to achieve collective objectives. For example, a group of agents might coordinate their paths to prevent collisions or their access to scarce resources to avoid conflicts. The focus is primarily on two approches: distributed constraint optimization (DCOPs) and auction-based decision-making.
In DCOPs, the collective decision problem is cast as a constraint optimization problem that is decentralized among the agents. A solution to such problems involves a protocol implemented by agents to coordinate their decisions. These protocols operate as distributed counterparts of conventional search or inference algorithms commonly employed in constraint processing.
In auction-based coordination, agents employ established auction mechanisms where collective decisions entail allocating tasks or resources to agents based on bids they place on sets of alternatives. Various auction schemes exist, each with distinct performance and operational characteristics, ranging from centralized combinatorial auctions to fully decentralized consensus-based auctions.
This presentation illustrates the applicability of these techniques through examples such as on-demand transportation, unmanned traffic management, and collective robotics.