Robbyant releases LingBot-Touch robot touch model

Robbyant, an embodied AI company within Ant Group, said Tuesday it had released LingBot-Touch as open source, describing it as a tactile foundation model for robots.

The company said the model converts magnetic tactile, visual tactile and piezoresistive signals into common tactile tokens for use by different robot hands and action models. It said LingBot-Touch was pre-trained on nearly 5,000 hours of real-world tactile data.

Robbyant said 26 real-world tests using two dexterous hands and four action models showed average task success-rate improvements of about 14 to 26 percentage points after LingBot-Touch was added. In one bottle-cap test, it said success rose from 56.25% without tactile input to 90% using the model's tactile representation.

The company said LingBot-Touch's technical report and code are available on Hugging Face and GitHub.