Reproduce “Tactile-Sensitive NewtonianVAE for High-Accuracy Industrial Connector Insertion” 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
Ryo Okumura, Nobuki Nishio, Tadahiro Taniguchi
A Tactile-Sensitive NewtonianVAE (TS-NVAE) is introduced to perform high-accuracy industrial connector insertion by integrating vision and tactile feedback for grasp pose compensation. The system achieved a 100% success rate with 0.3 at 0.1 mm positioning accuracy on a physical USB connector insertion task.
Key insights
- Inducing readily available domain knowledge (robot transition uncertainty, grasp pose variation) into the model's priors is critical for achieving high accuracy and robust learning in physical robot manipulation tasks.
- Directly mapping raw tactile sensor output to a grasp-compensated insertion position within a learned latent space bypasses complex calibration, explicit coordinate transformations, and annotation errors inherent in traditional multi-stage methods.
- World model frameworks can be effectively adapted for high-accuracy physical industrial manipulation by intelligently integrating multi-modal sensing and leveraging structured priors, leading to more practical and sample-efficient solutions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.