Paper releases LingBot-Touch tactile encoder weights and code

A paper proposes that how tactile data is represented may be more critical than resolution. The authors pretrained a tactile encoder, LingBot-Touch, on various sensors ranging from a 3x3 glove array to a 240x240 optical fingertip, and tested it with 4 strategies on 26 real tasks, while releasing the weights and code. The paper states that raw tactile signals hindered manipulation in 22 of 104 test groups, and that pretrained features could fix most of these. The authors designed all 26 tasks and the scoring criteria themselves, and the tests involved only two hand types.

AI 根据报道生成 · 1 days ago更新