🤖Tactile Datasets Help Robots Master Fiddly Tasks

Robots are getting a better sense of touch

Tactile datasets like T-Rex are revolutionizing how robots handle delicate tasks. These datasets, which include over 30,000 hours of synchronized visual and tactile data, are training models to predict tactile feedback and guide actions in real time. This is a big deal for anyone working on robotic manipulation, as it nearly doubles the success rate of complex tasks like screwing in a light bulb or applying toothpaste. The model, fine-tuned on 100 teleoperated demonstrations, averaged a 65% success rate across 12 tasks. However, the data is hardware-specific, and more research is needed to generalize tactile knowledge across different robot hands.