Contact-Anchored Policies: Contact Conditioning Creates Strong Robot Utility Models

2026-02-09

AI Summary

Language instructions are often too abstract to support robust robotic manipulation in complex physical interactions. To address this limitation, this work proposes replacing linguistic commands with spatial contact points as the conditioning signal for policy learning, constructing a modular utility model library, and integrating it with EgoGym—a lightweight simulation platform—to enable rapid real-to-sim closed-loop iteration. Using only 23 hours of demonstration data, the approach achieves out-of-the-box, zero-shot generalization across environments and robot embodiments on three fundamental manipulation tasks, outperforming state-of-the-art vision-language-action models by 56% in performance. The core innovation lies in the novel contact-point-conditioned policy architecture, which substantially enhances both generalization capability and deployment efficiency.

Technology Category

Intelligent Robots: ManipulationNatural Language Processing: Safety and RobustnessComputer Vision: Language and Vision

Application Category

Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactionsUser Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systemsEconomics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

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), 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 a 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. Project page: https://cap-policy.github.io/

Problem

• robot learning
• language conditioning
• physical manipulation
• generalization
• contact understanding

Innovation

• Contact-Anchored Policies
• modular utility models
• real-to-sim iteration
• physical contact conditioning
• EgoGym