Hierarchical Learned Auctions for Communication-Sparse Satellite Environments
Hannah C. Lehman, John Valasek · 2025
Multiagent coordination research tends to focus on strategies such as Multiagent Reinforcement Learning which makes extensive use of inter-agent communication. This works well in abstracted simulations, but may be challenging or non-feasible to implement in real-world environments because many problems of interest can have sparse, intermittent, or contested communication frameworks. This paper extends the Hierarchical Auctions for the Coordination of Heterogeneous Agents (HACHA) approach to investigate hierarchical auctions and machine learning to facilitate teaming between agents in a system where some agents enter and leave the communication graph at regular intervals, akin to a satellite. Results presented in the paper show that the method is able to identify, distribute, and monitor tasks despite communication challenges. This paper shows that allowing multiple children to propagate auctions results in lower average search times and higher success rates.