Search papers, labs, and topics across Lattice.
This paper introduces SemAnCorr, a novel framework that enhances zero-shot manipulation skill transfer by establishing dense correspondences through semantically consistent anchor regions. By optimizing joint pose-correspondence and utilizing functional maps, SemAnCorr achieves high semantic accuracy and improved geometric coherence, outperforming existing methods. The framework's effectiveness is demonstrated through real-world applications, where it significantly enhances the reliability of skill transfer to unseen objects using only a single demonstration.
Achieving 90.8% semantic accuracy, SemAnCorr revolutionizes zero-shot manipulation by enabling reliable skill transfer across diverse object geometries with just one demonstration.
Transferring manipulation skills across object instances that share functionality but differ in geometry remains a fundamental challenge in robot learning. While recent correspondence methods leverage dense visual descriptors and 3D feature fields, nearest-neighbor feature matching often produces spatially incoherent correspondences that fail to recover the local geometric frames required for reliable skill transfer. We introduce SemAnCorr, a training-free framework that establishes dense correspondence by selecting semantically consistent anchor regions through joint pose-correspondence optimization and propagating these constraints over the object surface using functional maps. The resulting correspondences preserve both semantic consistency and geometric coherence, enabling object-centric manipulation skills to transfer across geometrically diverse instances. We evaluate SemAnCorr on a dense correspondence benchmark built on PartNet-Mobility, achieving 90.8% semantic accuracy in our benchmark evaluation while improving geometric coherence over recent state-of-the-art baselines. Finally, we show that these improvements translate directly into real-world manipulation performance: using a single demonstration, SemAnCorr enables substantially more reliable zero-shot manipulation skill transfer to previously unseen objects than existing correspondence methods. Videos and additional visualizations are available at [https://semancorr.github.io](https://semancorr.github.io) .
