이영운 연구실

Generalizable skills from large-scale data and unsupervised interactions are essential for fast learning of new tasks. We strive for acquiring performant VLAs.

Skill-based learning extends the capability of robot learning to tasks with a much longer horizon via skill composition and hierarchical exploration.

Model-based RL and world models are key to achieve sample efficiency, especially with large-scale unsupervised interactions and diverse unlabeled data.

Whole-body manipulation demands breakthroughs to handle high-dimensional state and action spaces.

Robot learning in the real world requires high sample efficiency, offline training, and safe exploration. This can be accelerated with the extensive use of large task-agnostic data and simulators.