New pose-aware model cuts tactile glove errors by 18%
Flexible sensors confound hand poses with touch. A new algorithm fixes that.
Tactile gloves digitize touch during hand-object interactions, but their soft, flexible sensors are sensitive to both contact forces and hand pose changes—creating pose-related artifacts (PRAs). These artifacts cause misdetections, especially at low forces, raising the minimum detectable force (MDF) of the glove.
To solve this, Tianhong Catherine Yu and colleagues propose a glove-agnostic algorithm that leverages hand pose information (already captured by many glove systems) to separate true contact forces from pose-induced sensor deformations. Their model augments standard tactile-to-force pipelines with a residual prediction branch. Tested on three different glove designs with 15 users, the approach reduces MDF by 10.4%, 12.2%, and 18.3% respectively, with consistent improvements across all metrics. The framework requires no hardware modifications, offering a practical path to making tactile gloves more reliable for data collection, dexterous manipulation, teleoperation, and learning from demonstration.
- Pose-related artifacts (PRAs) cause tactile gloves to misdetect low-force contacts, raising minimum detectable force (MDF) by up to 18%.
- New pose-aware residual prediction branch estimates force corrections from hand pose data, reducing MDF by 10.4–18.3% across three glove designs.
- Algorithm is glove-agnostic and works without hardware changes, leveraging existing hand pose sensing.
Why It Matters
Makes tactile gloves usable for fine manipulation in robotics, teleoperation, and data collection without costly hardware redesigns.