The topic “Towards On-Premise Foundational Behavior Models” explored how recent breakthroughs in generative AI could be translated into practical, deployable intelligence for robotic systems. While diffusion models and vision-language models promised powerful, generalizable robot behavior, existing approaches were often too large, slow, data-hungry, and privacy-sensitive for real-world use. In this webinar, Rudolf Lioutikov presented recent advances that dramatically reduced model size, data requirements, and training and inference costs—spanning score-based diffusion, vision-language-action models, and action tokenization—charting a clear path toward efficient, privacy-preserving, on-premise foundational behavior models for robotics.
.jpg)
