로봇학습 연구실

Funded by the Ministry of Trade, Industry and Energy (MOTIE). The goal of this project is to develop a generalizable action generation framework that enables heterogeneous humanoid robots to perform complex, long-horizon tasks across diverse environments. We develop vision-language-action models that integrate perception, language instruction, locomotion, and manipulation, allowing robots with different morphologies to share common task representations and adapt learned behaviors across platforms. The project addresses key challenges in robot learning, including data scarcity, failure detection, trajectory correction, and transfer between simulation and real-world humanoids. By learning shared latent representations, task-topology structures, and compositional action sequences, the system can combine basic locomotion-manipulation skills into complex tasks such as object handling, preparation, and serving. The outcome will be a robust behavior generation engine and a data-sharing framework for heterogeneous humanoid robot intelligence. Funded by the Defense Acquisition Program Administration (DAPA). Funded by the Ministry of Science, ICT, and Future Planning (MSIP).