Reproduce “TACTIC: Tactile and Vision Conditioned Contact-Centric Control for Whole-Arm 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
Rishabh Madan, Angchen Xie, Samantha Saak, Andres Blanco, Dohyeok Lee, Sarah Grace Brown, Yunting Yan, Mark Zolotas and 2 more
TACTIC, a control framework, enables robots to execute robust and safe whole-arm manipulation by integrating multimodal sensing with learned and analytical predictive models. It achieves higher success rates and reduces unsafe force interactions in both simulated and real-world contact-rich tasks, including manikin caregiving.
Key insights
- A hybrid predictive model, combining learned latent dynamics with analytical kinematics, effectively balances physical consistency with the capacity to predict complex, multimodal interactions in whole-arm manipulation.
- Fusing distributed tactile sensing with a visual proximity mask is crucial for overcoming partial observability, significantly enhancing task success and safety in contact-rich scenarios.
- Contact-aware action sampling, which biases exploration towards force-modulating directions, substantially improves planning efficiency and reduces unsafe force violations in multi-contact environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.