Developing a novel reinforcement learning algorithm that allows robots to learn from sparse action-labeled data, under Professor Chelsea Finn.

Under the supervision of PhD student Suraj Nair, I am developing an algorithm in Python that will enable robots to learn to generalize to new environments given non-expert action-labeled data from other environments, and supplementing with expert action-free data from human demonstrations.

We are also collaborating with Sergey Levine and Ben Eysenbach at the Berkley RAIL Lab.