English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Humans excel at transferring manipulation skills across diverse objectshapes, poses, and appearances due to their understanding of semanticcorrespondences between different instances. To endow robots with a similarhigh-level understanding, we develop a Distilled Feature Field (DFF) for 3Dscenes, leveraging large 2D vision models to distill semantic features frommultiview images. While current research demonstrates advanced performance inreconstructing DFFs from dense views, the development of learning a DFF fromsparse views is relatively nascent, despite its prevalence in numerousmanipulation tasks with fixed cameras. In this work, we introduce SparseDFF, anovel method for acquiring view-consistent 3D DFFs from sparse RGBDobservations, enabling one-shot learning of dexterous manipulations that aretransferable to novel scenes. Specifically, we map the image features to the 3Dpoint cloud, allowing for propagation across the 3D space to establish a densefeature field. At the core of SparseDFF is a lightweight feature refinementnetwork, optimized with a contrastive loss between pairwise views afterback-projecting the image features onto the 3D point cloud. Additionally, weimplement a point-pruning mechanism to augment feature continuity within eachlocal neighborhood. By establishing coherent feature fields on both source andtarget scenes, we devise an energy function that facilitates the minimizationof feature discrepancies w.r.t. the end-effector parameters between thedemonstration and the target manipulation. We evaluate our approach using adexterous hand, mastering real-world manipulations on both rigid and deformableobjects, and showcase robust generalization in the face of object andscene-context variations.

SparseDFF: Sparse-View Feature Distillation for One-Shot Dexterous Manipulation