Behavior-Aware Arbitration for Scalable Robot Swarms in Cooperative Search-and-Retrieval Tasks

Open Access
Article
Conference Proceedings
Authors: Bradley Killen, Daniel Carruth, Cindy Bethel, Kenna Henkel, Victor Paul
Abstract

Adding agents to a team can improve search speed and task coverage, but larger teams can also increase coordination load, resource contention, and task interference. This paper studies robot-only swarm behavior as a step toward arbitration support for human-autonomy teaming. We developed a maze-based simulation in which agent teams collect keys, open doors, respond to coordination requests, and attempt to reach a shared objective. Agent behaviors include normal task execution as well as error-prone conditions that produce confused key handling, request overcommitment, task abandonment, or adversarial interference. Structured event logs were parsed into run-level, agent-level, request-level, and decision-level summaries. Experiments across multiple map sizes and team sizes show nonlinear performance scaling. Small maps were nearly always solved across team sizes, while larger maps were more often solved by mid-sized teams and showed degraded performance for larger swarm sizes. Request expirations and timeout-related events increased in higher-load conditions, suggesting that coordination overhead can reduce the value of additional agents. We also trained a contextual candidate-action arbitration model that converts event-stream features into supervisory recommendations, including door-progress prioritization, timeout-risk reassignment, continued exploration, key escort, and reduced reliance on agents associated with disruptive behaviors. The model is intended to centralize swarm activity and highlight when task progress, reassignment, or additional supervision may be needed. These findings support behavior-aware arbitration as a decision-support layer for identifying when larger autonomous teams are improving task progress and when they are increasing coordination burden.

Keywords: Robot Swarms, Adaptive Arbitration, Multi-Agent Systems, Autonomous Systems, Human-Autonomy Teaming

DOI: 10.54941/ahfe1008393

Cite this paper
Downloads
0
Visits
1
Download PDF

More from this volume

← An Intelligent Scheduling Capability for ISR Collection AssignmentA High-Availability Multi-Agent RAG Architecture for Automated Expert Assistance and Multimodal Technical Onboarding →
View all articles in Human Factors in Design, Engineering, and Computing →