I aim to develop scalable reinforcement learning (RL) algorithms that (1) improve performance as more data becomes available, (2) maintain robust performance in long-horizon episodes, and (3) generalize well across diverse tasks.

To achieve this, I believe it is essential to leverage the previously collected data and advance multi-task learning. With this perspective, I have been deeply engaged in data-driven RL, particularly offline RL. Additionally, I am interested in learning approaches such as preference-based RL, goal-conditioned RL, and imitation learning.