Date of Award:

5-2026

Document Type:

Dissertation

Degree Name:

Doctor of Philosophy (PhD)

Department:

Computer Science

Committee Chair(s)

Mario Harper

Committee

Mario Harper

Committee

John Edwards

Committee

Soukaina Filali Boubrahimi

Committee

Greg Droge

Committee

Jacob Gunther

Abstract

Autonomous robot teams must move reliably, explain their actions, and work together—even in changing or adversarial settings. We present a structured, three-stage pathway that builds coordinated team behavior from strong single-robot skills. First, we develop robust legged-robot locomotion and interpretability using only internal actuator signals. From these proprioceptive cues, robots learn to classify terrain and predict short-term power use, enabling energy-aware movement without external sensors. We further analyze learned behaviors with motif discovery to reveal recurring sensor–action patterns, which clarify how agility emerges and guide reward design. Second, we compose these skills into heterogeneous teamwork using centralized training with decentralized execution. Role-conditioned critics and attention mechanisms allow robots with different bodies to share workload and cooperate. We evaluate cooperative transport, mixed-morphology block pushing, precise hand-to-hand transfer, and exploration in the Unknown Building Exploration Simulator, measuring success, efficiency, and resilience to agent dropouts. Third, we introduce adversarial multi-team learning in IsaacLab with a scalable, physics-faithful stack. Self-play and population-based curricula harden policies against opponents, while team-specific critics and a zero-buffer curriculum stabilize training as tasks grow more complex. Competitive tests include Sumo, Soccer, and a 3D “Galaga” interception scenario, reported by win rate. Together, these stages connect interpretable, energy-aware single-robot skills to coordinated, resilient multi-robot behavior, yielding a GPU-accelerated, reproducible framework for heterogeneous teams in adversarial environments.

Download

Share

COinS

Copyright for this work is retained by the student. If you have any questions regarding the inclusion of this work in the Digital Commons, please email us at .