This work package has 3 main research areas: Cooperative game theory and learning; Human-robot interaction; Heterogeneous robot-robot interaction and reinforcement learning algorithms
Between them they are exploring learning in open multi-agent systems, using game-theoretic approaches to develop multi-agent (reinforcement) learning and trustworthiness.
Outputs to date include a NeurIPS 2025 article showcasing work around Shapley values and their use in multi-agent reinforcement learning Shapley Machine: A Game-Theoretic Framework for N-Agent Ad Hoc Teamwork.
Using multimodal sensing adapted to robot behaviour, the team have also identified intricate connections between 3D eye-gaze tracking and subjective trust in humans as well as finding ways to capture the uncertainty of predicting human navigational behaviour.
In addition, building on previous work with colleagues at University of Cambridge . some key technical contributions have been identified to measure diversity and analyse its impacts in multi-agent learning; as highlighted in the work ‘System Neural Diversity’ . [2305.02128] System Neural Diversity: Measuring Behavioral Heterogeneity in Multi-Agent Learning
