Reproduce “TactiDex: A Real-World Tactile-Guided Benchmark for Human-Like Dexterous Manipulation” 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

Suting Ni, Hanbing Zhang, Zhenyu Wei, Guo Chen, Chixuan Zhang, Ye Shi, Jingya Wang

TactiDex introduces a real-world, tactile-guided benchmark for dexterous manipulation, while TactiSkill is a framework that leverages synchronized whole-hand tactile signals as structured supervision. This approach achieved an 81.95% kinematic success rate and a 64.64% tactile-aware success rate, significantly improving physical fidelity and safety in robotic manipulation tasks over kinematic-only methods.

Key insights

  • Integrating synchronized whole-hand tactile feedback as structured supervision is vital for progressing from kinematic imitation to physically grounded, human-like dexterous manipulation.
  • A composite tactile reward, which encourages contact formation, aligns simulated forces with human patterns, and penalizes excessive forces, is essential for stable and safe contact-rich manipulation.
  • Policies trained with tactile supervision in simulation can achieve robust, physics-aware force distributions that enable effective zero-shot sim-to-real transfer, even without real-time tactile feedback on the physical robot.

Summary by alphaXiv.

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