Welcome to my homepage!
My research focuses on reinforcement learning (RL), and its applications in robotics. Recent work includes investigating deep neural network structures that are suitable for learning to plan and control, such as the value iteration network. Another example is our work on learning cost shaping for model predictive control using a hindsight calculation.
Previous work focused on various risk-sensitive performance criteria, as opposed to the conventional expected total reward criterion. In particular, we have shown that many classic RL approaches may be extended to risk-sensitive criteria such as mean-variance optimization and conditional value-at-risk (CVaR).See my publications page for more details.
11/2016: Our paper titled Value Iteration Networks won the NIPS best paper award!