@article{cui2026contact,
title={Contact-Anchored Policies: Contact Conditioning Creates Strong Robot Utility Models},
author={Zichen Jeff Cui and Omar Rayyan and Haritheja Etukuru and Bowen Tan and Zavier Andrianarivo and Zicheng Teng and Yihang Zhou and Krish Mehta and Nicholas Wojno and Kevin Yuanbo Wu and Manan H Anjaria and Ziyuan Wu and Manrong Mao and Guangxun Zhang and Binit Shah and Yejin Kim and Soumith Chintala and Lerrel Pinto and Nur Muhammad Mahi Shafiullah},
journal={arXiv preprint arXiv:2602.09017},
year={2026}
}
Contact-Anchored Policies
Contact Conditioning Creates Strong Robot Utility Models
The prevalent paradigm in robot learning attempts to generalize across environments, embodiments, and tasks with language prompts at runtime. A fundamental tension limits this approach: language is often too abstract to guide the concrete physical understanding required for robust manipulation. In this work, we introduce Contact-Anchored Policies (CAP), a new class of general robotic behavior models, which replace language conditioning with points of physical contact in space. Simultaneously, we structure CAP as a library of modular utility models rather than a monolithic generalist policy. This factorization allows us to implement an efficient real-to-sim iteration cycle: we build EgoGym, a lightweight simulation benchmark, to rapidly identify failure modes and refine our models and datasets prior to real-world deployment.
We show that by conditioning on contact and iterating via simulation, CAP generalizes to novel environments and embodiments out of the box on three fundamental manipulation skills while using only 23 hours of demonstration data, and outperforms large, state-of-the-art VLAs in zero-shot evaluations by 56%. All model checkpoints, codebase, hardware, simulation, and datasets will be open-sourced.
Contact-anchored policies (CAPs) are trained on a diverse dataset of
contact-anchored
interactions:
Contact Anchoring
Deployment across embodiments in real and sim
