MatchingPolicy: Correspondence-Aware Policy for Cross-Object In-Context Learning

1University of XXX,

Abstract

In-context imitation learning allows policies to generalize from few-shot demonstrations, yet it often struggles with unseen objects or novel scenarios. To address this, we introduce MatchingPolicy, a correspondence-driven framework that decouples demonstration-to-scene matching from policy learning. At its core, MatchingPolicy uses a graph-based diffusion policy that adapts robot actions via dense semantic correspondences. This explicit separation eliminates the challenge of simultaneous correspondence inference and action adaptation, enabling robust transfer. Our approach integrates a vision foundation model with a novel two-stage matching algorithm to dynamically establish reliable correspondences. Empirical evaluations on RLBench and real-world manipulation tasks show that MatchingPolicy achieves strong few-shot performance, generalizing consistently across unseen object instances and categories.

Method Overview

Overall Framework of MatchingPolicy. The overview of MatchingPolicy. Correspondence points are extracted using a two-stage feature matching algorithm and passed into a graph-based diffusion policy model, which predicts the 6D motions of a gripper. These motions are subsequently converted into robot joint commands.

Visualization of Real-world Deployment

 

Human Demo

Layout Generalization

Object Generalization


Open Box
Pour Water
Put Lid
Cut Egg

Multi-Object & Distractors

Evaluation in cluttered scenes with distractors


Human Demo

Policy Rollout

Visualization of Different Matching Algorithms

 

Naive 3D Matching

FPFH (Fast Point Feature Histograms)

Two-Stage Matching (Ours)


Open Box
(Large Layout Diff.)
Pour Water
(Large Layout Diff.)
Open Box
(Shape Diff.)
Pour Water
(Shape Diff.)

Failure Analysis: InstantPolicy


Failure Case of InstantPolicy: Here, the baseline fails to interact with the target object correctly because it relies on implicit geometric features which are ambiguous in this layout.

Visualization of Simulation Deployment

Phone on Base

Umbrella Out


More results...


BibTeX

@article{matchingpolicy,
  author    = {XXX},
  title     = {MatchingPolicy: Correspondence-Aware Policy for Cross-Object In-Context Learning},
  journal   = {xxx},
  year      = {xxx},
}