Robust Autonomy and Decisions Group

We develop machine learning methods that enable autonomous robots to operate robustly in application domains including the following:

  • 'Programming by discussion', to teach autonomous robots to perform complex tasks, e.g., surgical operations and complex inspection procedures
  • Active sensing, predictive modelling and decision making, e.g., in medical diagnostics
  • Human-robot collaborative work such as intention-aware assistance in joint manipulation tasks

These applications motivate us to develop new models and algorithms, such as:

  • Compositional and incremental methods for robust model learning and learning to act safely
  • Learning causal and interpretable generative models of dynamical processes, including techniques such as probabilistic programming
  • Mechanisms of extended human-AI interaction for model selection and structure learning

News feed