Rhoda AI pioneers direct video action models to train robots from internet video data, addressing robotics industry's data shortage challenge

Rhoda AI, co-founded by Eric Chan (a researcher with prior roles at NVIDIA, Google, NASA, and WorldLabs), emerged from stealth in March 2026. Chan serves as co-founder and chief scientist and leads research on foundation models for physical AI at the Palo Alto, Calif.-based company.

The company uses direct video action (DVA) models—systems that train robots using internet video data—to enable efficient data use and allow robots to tackle complex tasks with minimal training, addressing the fundamental problem of insufficient training data for physical AI companies.

Chan discusses challenges of data collection, potential for zero-shot learning (learning without task-specific examples), and the future of robot deployment in real-world applications in Episode 242 of The Robot Report Podcast.