My research focuses on applying Safe Reinforcement Learning to Behavioural Planning in Autonomous Driving.
In other words, I address the problem of how to take high-level driving decisions (e.g. lane changes) - in the presence of uncertainty (e.g. in the behaviour of other traffic participants).
Driving is a social game: interactions between drivers introduce a coupling between the decisions of the ego-vehicle and the dynamics of the nearby traffic. They are difficult to model manually and justify the need to learn from data. But to deploy a learning-based system in such a critical setting, a notion of risk must be considered explicitly and controlled.