Reproduce “Object Recognition and Force Estimation with the GelSight Baby Fin Ray” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Sandra Q. Liu, Yuxiang Ma, Edward H. Adelson

Researchers at MIT demonstrated that the GelSight Baby Fin Ray, a compact, camera-based tactile sensor integrated into a soft robotic finger, can simultaneously achieve 97.5% accuracy in texture classification and precisely estimate contact position (2.02 mm error) and normal force (1.18 N error) using deep learning. The system extracts multiple modalities of information from a single sensor.

Key insights

  • A single, compact vision-based tactile sensor can effectively provide rich, multi-modal information including texture, applied force, and contact position for soft robotic fingers.
  • Deep convolutional neural networks, particularly GoogLeNet, excel at extracting complex tactile features from raw visual data, outperforming shallower networks and traditional methods for both classification and regression tasks.
  • Integrating high-resolution tactile sensing with compliant soft robotics, enabled by machine learning, enhances a robot's ability to understand and skillfully interact with diverse objects and environments.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.