MScE Defence - Matthew Tidd - Department of Mechanical Engineering - FR

Event date(s): October 13, 2026
Time(s): 09:00 AM - 11:00 AM
Category: Fredericton
Location: Fredericton

ABSTRACT:  Coordinating heterogeneous multi-robot systems requires effective task allocation and reliable task execution, despite differences among agents within the system, as well as the ability to handle isolated single-agent failures. This thesis explores the intersection  between deep reinforcement learning, behaviour trees, and classical multi-robot task allocation methods to form a fully autonomous pipeline for facilitating task detection, distributed allocation, and goal-directed execution for use in heterogeneous multi-robot systems. A common behaviour tree manages task reception, distributed capability-aware task auctioning, task navigation using embedded agent-specific DRL policies, and failure-contingent re-auctioning. The feasibility of this methodology is validated both in simulation and through practical testing. Across 25 simulated missions comprising 200 task allocations and executions, the system completed every task, balanced workload among agents, and recovered from navigation failures through re-auctioning. Conversely, 36 tasks across six physical missions were completed without collision, timeout, or re-auctioning. 

Building: Head Hall

Room Number: TME Room 224