M1 Room

Models of Human Behaviour

M1 develops formal and computational models for predicting complex goal-oriented actions and multi-agent collaborations across time scales to advance Human-in-the-Loop computations for digital technologies.

Key Questions

We will extend models to link continuous movements with discrete decisions.

We will develop collaborative models by extending POMDPs to include multiple agents.

We will use contextual reinforcement learning to model complex behavior acquisition.

Approach

  • Sensorimotor Sequences: Extend current models to multi-joint movements, validated with motion capture and brain imaging data.
  • Collaborative Actions: Extend POMDPs to represent state transitions in joint actions, in collaboration with F1, M2, and U1.
  • Skill Learning: Detect errors and provide real-time feedback during skill acquisition, tested in surgery (U1).

Expected Results

  • Full specification of hierarchical models and collaborative action models.
  • Integration of contextual inference to simulate expert performance.