Tactile-JEPA: Topology-Aware Self-Supervised Representation Learning for Distributed Tactile Sensors
Tactile-JEPA pretrains topology-aware encoders for sparse robot tactile skins using graph-guided masking.
Tactile-JEPA is a self-supervised pre-training method for distributed electronic skins whose sensing elements are sparse and irregular. It predicts embeddings of masked elements from the remainder, using the sensor connectivity graph and dual-scale masking to capture local contact and global surface state. Across three datasets with magnetic and piezoresistive sensors and different robot embodiments, it reduces force estimation error by 6.3% and in-hand orientation error by 20.8% versus prior state of the art, with gains in policy learning.
- Graph-guided masking respects irregular tactile sensor layouts
- Dual-scale masking captures local contact and global surface state
- Force error drops 6.3% and orientation error 20.8%
- Tested on magnetic and piezoresistive skins across embodiments
From the original article
1 min read at huggingface.co
Tactile sensing is an essential modality for robots performing contact-rich, dexterous manipulation, particularly under visual occlusion. While pre-trained image encoders are standard in robot learning pipelines, tactile encoders are still commonly trained from scratch from raw, noisy signals, which might limit their expressivity.
By Elizaveta Kovtun, Matvey Konovalov, Andrey Sakhovskiy · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.