Imitation Learning from human teleoperation has proven effective for training robots in dexterous, high-dimensional tasks. At the same time, learning directly from video demonstrations is emerging as a complementary approach, offering scalable data sources extending beyond traditional teleoperation. With the rise of large pre-trained models in robotics, a key frontier is understanding how both teleoperated and video-based demonstrations can fine-tune these models for diverse downstream applications. This workshop will explore emerging challenges and opportunities at this intersection: How should teleoperation and video data be collected, structured, and scaled to maximize transfer to large models? What role do modalities such as tactile sensing, haptics, and multimodal feedback play in enabling robust fine-tuning? And how can we benchmark and standardize practices across data sources and teleoperation setups to accelerate progress?

Building on the success of the first edition, this workshop will convene leading researchers to present perspectives on leveraging human demonstrations — from teleoperation to video — for large-model adaptation. Through invited talks, a poster session, and lightning presentations, participants will engage in discussions on dataset design, evaluation protocols, and integration of new sensing modalities. The goal is to foster collaboration across the community to establish best practices.