Timezone: »

Robust Reinforcement Learning using Offline Data
Kishan Panaganti · Zaiyan Xu · Dileep Kalathil · Mohammad Ghavamzadeh

Tue Nov 29 02:00 PM -- 04:00 PM (PST) @ Hall J #509

The goal of robust reinforcement learning (RL) is to learn a policy that is robust against the uncertainty in model parameters. Parameter uncertainty commonly occurs in many real-world RL applications due to simulator modeling errors, changes in the real-world system dynamics over time, and adversarial disturbances. Robust RL is typically formulated as a max-min problem, where the objective is to learn the policy that maximizes the value against the worst possible models that lie in an uncertainty set. In this work, we propose a robust RL algorithm called Robust Fitted Q-Iteration (RFQI), which uses only an offline dataset to learn the optimal robust policy. Robust RL with offline data is significantly more challenging than its non-robust counterpart because of the minimization over all models present in the robust Bellman operator. This poses challenges in offline data collection, optimization over the models, and unbiased estimation. In this work, we propose a systematic approach to overcome these challenges, resulting in our RFQI algorithm. We prove that RFQI learns a near-optimal robust policy under standard assumptions and demonstrate its superior performance on standard benchmark problems.

Author Information

Kishan Panaganti (TAMU)
Kishan Panaganti

I am a fifth-year Electrical and Computer Engineering PhD student at Texas A&M University, where I have the privilege of being advised by Prof. Dileep Kalathil. I work on reinforcement learning algorithms that span several areas like optimization, high-dimensional probability, multi-armed bandits, and stochastic theory (this is a non-exhaustive list). My near-future plans will be to contribute theoretical guarantees for the algorithms under safe and robust regimes of reinforcement learning. I am also open for post-doc job market starting Summer 2023!

Zaiyan Xu (Texas A&M University)
Dileep Kalathil (Texas A&M University)
Mohammad Ghavamzadeh (Google Research)

More from the Same Authors