Research Software Engineer

Google

In this work, we address the practical problem of training MPC policies with parameterised cost functions when the demonstrator and the imitator agents have non-identical state-action spaces and transition dynamics. We propose a novel approach that uses a generative adversarial network (GAN) to match state-trajectory distributions of the demonstrator and the imitator. We demonstrate its efficacy on MuJoCo tasks of the DeepMind Control suite.

In this work, we propose a modular framework that is able to efficiently search real indoor environments for objects (e.g. fruits, glasses, phones, etc.) that frequently change their positions due to human interaction. We train a Contextual Bandit agent for exploration and leverage Weighted Minimum Latency Solvers and Learnable MPC motion planners for high sample efficiency, safety and reliability.

Motivated by practical problems of ranking with partial information, we introduce a variant of the cascading bandit model that considers flexible length sequences with varying rewards and losses.  Our analysis delivers tight regret bounds which, when specialized to standard cascading bandits, results in sharper guarantees than previously available in the literature.