Tactile gloves are designed to capture contact and force during hand-object interactions, enabling advances in dexterous manipulation, teleoperation, and learning from demonstration. However, the soft and flexible nature of these gloves introduces pose-related artifacts that corrupt the tactile data.
These artifacts arise from variations in hand pose, such as finger bending and wrist orientation, which alter the sensor response even under the same contact conditions. This poses a significant challenge for downstream tasks that rely on consistent tactile feedback.
To address this issue, the authors propose a pose-aware modeling approach that explicitly accounts for hand pose during tactile data acquisition. The method integrates pose information into the sensor model to correct for artifacts, improving the fidelity of the tactile signals.
Experiments demonstrate that the proposed approach effectively reduces pose-related distortions, leading to more reliable tactile data. This enhancement is critical for applications like robotic manipulation, where precise force and contact sensing is required.
Source: arXiv (2607.22964).