Humans and Autonomous Agents Lab

Safety of Autonomous Vehicles and Agent-Based Systems

As autonomous agents—such as self-driving vehicles and intelligent assistants—become more widespread, ensuring their safety and reliability in human-centered environments is critical. Our work explores novel approaches to verification and validation that explicitly account for human unpredictability, cognitive diversity, and interaction dynamics. We aim to develop methodological frameworks that address both technical correctness and behavioral safety.

Game theory for Human-AI interaction

We focus on human-AI interaction from a behavioural game theoretic lens. We study how to formally represent interactions between humans and autonomous agents, especially when human behavior deviates from classical models of full rationality. Drawing from behavioral and empirical game theory, we develop models of bounded rationality to better align automated decision-making with real-world human behavior. This line of work also examines the strategic implications and risks that arise when agents rely on incorrect or idealized models of human decision-making.

Cognitive architectures for Generative AI agents

Design and development of generative AI based agents with a focus on cognitive architectures that support learning in open-world systems

Social and Economic Modeling with AI-based simulation

We leverage large language models and formal simulation frameworks to model human behavior at both micro and macro levels. This research develops tools to study social, economic, and institutional dynamics through AI-based simulations, thereby supporting the testing of hypotheses and the design of interventions in complex, societal-scale systems.

Norms, Trust, and Values in Agent-based AI systems

This line of research deals with reasoning, learning, and deliberation of norms within multiagent systems. The overarching goal is to look beyond preferences (stated or revealed) and move towards a more holistic sense of autonomy.