Panasonic Holdings and Panasonic R&D Company of America have developed T-Rex, a robot manipulation AI that uses touch at the fingertips to adjust two-handed tasks in real time, together with UC Berkeley, NVIDIA, Stanford University, Sapienza University of Rome and ItalAI, iRobotNews reports. Panasonic announced it on 5 November.
T-Rex adds fingertip touch data to a vision-language-action model. One part plans movements from camera images and language about five times a second, while another reads contact about 20 times a second to correct for slipping or too much pressure without running image recognition again.
The team collected more than 7,700 two-handed robot demonstrations, about 100 hours, with touch sensors on ten fingertips. In 12 real tasks such as turning pages, moving eggs and opening a lock with a key, T-Rex averaged a 65% success rate, against 35% for the best model it was compared with. The work is to be presented at CoRL 2026 in Austin from 9 to 12 November.
