Robotics: why now? - Quan Vuong and Jost Tobias Springberg, Physical Intelligence

Quan and Toby discuss their mission to create a universal robot control model, highlighting the limitations of current robotics in unstructured environments and the advancements made possible by AI and vision language action models (VLAs). They detail the engineering challenges in VLA training, particularly data sourcing and model deployment, and explain their approach to building a data engine using human-operated robots and cloud-based annotation. They introduce PIO5, a VLA with open-world generalization, demonstrating its ability to perform long-horizon tasks in unseen environments, and emphasize the importance of diverse data collection. They are also seeking partnerships and talent to help accelerate progress towards their mission.