A robotics study involving computer vision expert Fei-Fei Li’s team, named T-Rex, was released in June and demonstrated that equipping robots with high-frequency tactile feedback enabled delicate tasks such as flipping book pages, handling eggs, squeezing toothpaste, and feeling mahjong tiles. Researchers collected 100 hours of robot manipulation data embedded with tactile signals, tasking robots with 12 operations. T-Rex achieved an average success rate of 65% across all 12 tasks, 30 percentage points higher than the strongest baseline model. The robotic platform used was the Dexmate Vega-1 dual-arm robot, equipped with two 22-degree-of-freedom Sharpa Wave dexterous hands, each fitted with five fingertip tactile sensors. The perception suite included a ZED X Mini stereo camera and two wrist-mounted cameras. The dataset combined 207 everyday objects with 22 basic actions, filtering down to 502 viable combinations, each recorded approximately 17 times, culminating in 100 hours of tactile-synchronized data. Success rates for individual tasks included flipping book pages at 96%, transferring eggs at 75%, wiping plates at 69%, squeezing toothpaste at 66%, separating stacked cups at 78%, feeling mahjong tiles at 65%, unlocking locks at 47%, dispensing pills at 41%, acid-base neutralization at 76%, extracting cards at 70%, dealing poker cards at 57%, and screwing in light bulbs at 35%. The research found