MULTI-ROBOT COORDINATION AND COOPERATION VIA GRAPH-BASED COMPUTATION

Loading...
Thumbnail Image
Degree type
Doctor of Philosophy (PhD)
Graduate group
Mechanical Engineering and Applied Mechanics
Discipline
Electrical Engineering
Subject
Machine learning
Multi-robot systems
Robotics
Task Planning
Funder
Grant number
License
Copyright date
01/01/2024
Distributor
Related resources
Author
Gosrich, Walker
Contributor
Abstract

Multi-robot coordination and cooperation are critical behaviors that improve team capabilities and enable new tasks in application areas like autonomous construction, agriculture, and extended operation in large unknown regions. This dissertation examines these behaviors in the context of the multi-robot resource allocation problem, where robots must be allocated to regions of service. In particular, we are interested in uncertainty-tolerant approaches that apply to large multi-robot teams. We introduce a graph-based modeling framework for the multi-robot resource allocation problem that offers unprecedented richness in representing inter-region relationships and reward models. We first address the multi-agent coverage control problem, introducing graph-based computation via Graph Neural Networks, which boasts improved performance and scalability by leveraging learned inter-agent communication strategies. We then address the multi-robot task allocation problem in complex multi-task missions where coordination and cooperation are explicitly required. We introduce a network-flow-based planning approach that produces high quality solutions to large problems in seconds. We expand this approach into an online setting that re-plans around task failures and unexpected observations. We demonstrate empirically that these modeling approaches and algorithms bring performance improvements that further the state of the art by leveraging the fundamental graph structure present in some multi-robot problems.

Advisor
Yim, Mark
Kumar, Vijay
Date of degree
2024
Date Range for Data Collection (Start Date)
Date Range for Data Collection (End Date)
Digital Object Identifier
Series name and number
Volume number
Issue number
Publisher
Publisher DOI
Journal Issue
Comments
Recommended citation