Seungho Baek

I'm a Ph.D. student at Sungkyunkwan University (SKKU), advised by Prof. Yusung Kim. I'm interested in developing reinforcement learning (rl) algorithms for long-horizon and sparse-reward tasks, and applying them to real-world robotic control. The question that drives me these days is: how can we effectively combine foundation models with rl? More specifically, my goal is to design off-policy q-learning methods that improve the long-horizon reasoning abilities of vision-language-action (vla) models.

My research interests include the following challenges:

• Discovering skills for generalization
• Learning robustly from suboptimal datasets
• Combining foundation models with off-policy rl
• Solving long-horizon tasks with sparse rewards

Research Keywords: Offline RL · Off-to-Online RL · Unsupervised RL · Goal-Conditioned RL · Hierarchical RL