Robotics labs are collecting motion and video data at scale to train the AI models that power their machines. Dyna Robotics demonstrated the effectiveness of this approach: its Dyna-2 model saw task success climb from 20% to 53% in August 2026 after expanding its training dataset from 1,000 hours to over 1 million hours of egocentric footage. Innodata opened a motion-capture R&D lab in Ridgefield Park, New Jersey, on September 30, 2026, while Agile Robots announced its Robot Academy in Munich around October 8, 2026. Robotics and physical AI companies have raised nearly $48 billion in funding in 2026, shifting from simulated datasets to large collections of human video and precision motion data. A 53% success rate still means nearly half of attempts fail, indicating significant challenges remain.
Source: Read the original article
