Towards an Adaptive System for Investigating Human-Agent Teaming: Let’s get Cooking!

Militaries will see an increased reliance on autonomous systems collaborating with human teams. Concepts of operations for these systems include humans in- or on-the-loop to ensure effectiveness and ethicality. Still, autonomy will require more than agents executing predefined tasks and be required to adapt to their human teammates. Experimental testbeds designed to study cooperative human-agent adaptation are lacking. This paper presents CHAOPT-2E: Cooking with Humans and Autonomy in Overcooked for studying Performance and Teaming: Eye tracking Edition. CHAOPT-2E is designed to 1) implement existing or new autonomous agents, 2) vary the environmental context, 3) obtain detailed human, agent, environment, and team data, 4) acquire multiple human inputs including eye gaze, and 5) provide capabilities for mutual adaptation experiments. To validate the system, we collected data from 81 participants in the CHAOPT-2E gaming environment with autonomous teammates. Test results showed that the system reliably collected data specific to humans, agents, environments, and teams, which was subsequently validated through multiple analyses. As such, this work provides foundational elements for future research aimed at incorporating human-agent information for mutual adaptation.