Multi-agent system for formation flying missions

Abstract

Concerns use of spacecraft as autonomous coordinated teams. Generalized reasoning capability offered by advanced distributed software technology and AI can cope with unexpected events and uncertainty, and so close the loop of perception, decision and eventually deliberation. The team members play interchangeable roles and negotiate about the task. We present a multi-agent system to provide a high degree of autonomy and support for coordination among team members. We use JPL formation flying mission initial architectures as benchmark. Our target is to avoid inconsistencies/disagreements between two or more participants in a collaborative context, increase the system's fault tolerance in cases such as loss of a member while the system still operates reliably. We address cooperation between collaborating independent autonomous agents. In a top-down organization agents are coordinated hierarchically, where the agents at the top of the hierarchy make the majority of the intelligent group decisions. In a more structured but still hierarchical organization, lower-level agents exercise more intelligence. A lower-level agent can advance a plan for the others to follow, and a higher-rank agent decides on the best plans. Although more rigid, the centralized intelligence organization allows for less communication among agents, so is more straightforward to implement. The decentralized approach requires more communication, but the intelligence is truly distributed, which makes for a more flexible, adaptive and efficient organization.