Robbyant's LingBot-Touch unifies three tactile sensor types
TL;DR
- LingBot-Touch learns a single representation across piezoresistive, magnetic, and visuotactile sensors, with the abstract calling the design extensible to new sensor types.
- The model outputs plug-and-play tactile tokens that feed into diffusion policy and VLA stacks including π₀.₅ and the group's own LingBot-VLA.
- The paper topped Hugging Face's daily list on November 10 with 80 upvotes, backed by simulation and real-robot runs on multiple hardware platforms.
Robbyant's new LingBot-Touch paper describes a tactile foundation model that learns a single representation across three distinct sensor families: piezoresistive, magnetic, and visuotactile. The preprint, titled "Learning Heterogeneous Tactile Representation for General Dexterous Manipulation," topped Hugging Face's paper-of-the-day list on November 10 with 80 upvotes.
The output the authors care about is what the abstract calls "plug-and-play tactile tokens that consistently enhance manipulation policies such as diffusion policy and VLA models," with π₀.₅ and the group's own LingBot-VLA named as the integration targets. The paper frames the broader claim as "a unified, extensible representation across heterogeneous tactile sensors," with the extensibility covering sensor types the team has not pretrained on.
Validation runs on simulation and real-robot deployments across multiple hardware platforms. The abstract publishes no per-sensor or per-task success numbers, and the GitHub repo at Robbyant/lingbot-touch sits at 66 stars a day after submission.
It lands in a heavy robotics week on our side too: the tracker logged 195 robotics stories over the last 90 days, with SJTU AutoLab's A²-MEAS benchmark run and MIT and NYU's FLEX both posting the same day.
Originally reported by huggingface.co
Read the original article →Original headline: Robbyant's LingBot-Touch Unifies Piezoresistive, Magnetic and Visuotactile Sensors in One Foundation Model