Distributed Multi-Agent Decision Making Under Uncertain Communication

Abstract

As space exploration accelerates and the number of robots and humans working in extreme environments grows with it, we must enact autonomous multi-agent coordination in order to safely operate in environments that are inherently hostile to communication. To the best of our knowledge, there are no multi-agent scheduling algorithms capable of independently reasoning over communication delay. A key gap that must be addressed is a single-agent scheduler capable of deciding when to act given uncertain observation, which can the form the basis for distributed multi-agent scheduling. Existing research has provided insights into temporal reasoning, namely modeling observation uncertainty and scheduling events with temporal constraints. There is both a need for deciding when to schedule events when there is uncertain observation delay, and a need to robustly coordinate between agents. Scheduling events in the face of uncertainty is a challenge due to the compounding uncertainties of uncontrollable exogenous events, unknown observation delay, and uncertain communication between agents. This thesis puts forth a series of contributions that culminates in the demonstration of a robust single-agent task executive that used our scheduler to coordinate in a multi-agent context despite observation delay. Doing so required insights in checking controllability of temporal constraints with uncertain delay, defining a scheduler that is robust to uncertain observation delay, integrating the scheduler in an existing high-level task executive, and a coordination strategy for multiple agents. We show that the scheduler exhibits the expected performance characteristics, and perform laboratory demonstrations of multi-agent execution with uncertain communication using a scenario inspired by human spaceflight.