Research
We focus on the research areas described below. This overview paper outlines some of the problems and challenges tackled by our research. For more information about our research, see People and Publications.
Multi-Agent Reinforcement Learning
We develop algorithms for multi-agent reinforcement learning to enable autonomous agents to communicate, coordinate, and collaborate in complex environments. Recent work also explores how large language models (LLMs), vision-language models (VLMs), and vision-language-action (VLA) models can be used to ground multi-agent interaction in rich reasoning, language, and visual understanding. A core challenge is to achieve convergence to high-performance equilibria and efficient scalability to large agent populations where joint decision spaces become intractable for centralised approaches, allowing agents to communicate and act optimally to achieve their goals.
Recent publications:
NashPG: A Policy Gradient Method with Iteratively Refined Regularization for Finding Nash Equilibria
Pareto Actor-Critic for Equilibrium Selection in Multi-Agent Reinforcement Learning
Benchmarking Multi-Agent Deep Reinforcement Learning Algorithms in Cooperative Tasks
Decision Making and Modelling Other Agents
Our long-term goal is to create autonomous agents capable of robust goal-directed interaction with other agents, with a particular focus on "ad-hoc" teamwork problems that require rapid and effective adaptation without prior coordination between agents. We develop algorithms enabling an agent to reason about the capabilities, behaviours, and intentions of other agents from limited observations, combining such inferences with reinforcement learning and planning for effective decision making. Increasingly, we leverage foundation models to equip agents with richer reasoning capabilities, allowing them to interpret context, anticipate the intentions of others, and interact robustly with previously unseen partners.
Recent publications:
From Autonomy to Alliance: Robotic Foundation Models Must Learn With Us, Not Just For Us
A General Learning Framework for Open Ad Hoc Teamwork Using Graph-based Policy Learning
Agent Modelling under Partial Observability for Deep Reinforcement Learning
Applied AI & Industry Collaborations
We collaborate with industry partners to develop real-world applications of AI, tackling the challenges that arise when autonomous systems must operate reliably in complex, dynamic, and safety-critical environments. Our applied work includes motion planning and prediction for autonomous driving, scalable coordination for multi-robot warehouses, and human-AI workflow orchestration — domains that demand robust perception, fast decision making, and seamless coordination between humans and machines. A central challenge across these applications is bridging the gap between controlled research settings and the unpredictability of real deployments, where systems must handle rare events, incomplete information, and strict operational constraints.
Recent publications:
Orchestrating Human-AI Teams: The Manager Agent as a Unifying Research Challenge
Scalable Multi-Agent Reinforcement Learning for Warehouse Logistics with Robotic and Human Co-Workers
Interpretable Goal-based Prediction and Planning for Autonomous Driving