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 Plus annual

Presents the access of premium content as premium feature

Premium Content

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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

We study how choices of input point cloud coordinate frames impact learningof manipulation skills from 3D point clouds. There exist a variety ofcoordinate frame choices to normalize captured robot-object-interaction pointclouds. We find that different frames have a profound effect on agent learningperformance, and the trend is similar across 3D backbone networks. Inparticular, the end-effector frame and the target-part frame achieve highertraining efficiency than the commonly used world frame and robot-base frame inmany tasks, intuitively because they provide helpful alignments among pointclouds across time steps and thus can simplify visual module learning.Moreover, the well-performing frames vary across tasks, and some tasks maybenefit from multiple frame candidates. We thus propose FrameMiners toadaptively select candidate frames and fuse their merits in a task-agnosticmanner. Experimentally, FrameMiners achieves on-par or significantly higherperformance than the best single-frame version on five fully physicalmanipulation tasks adapted from ManiSkill and OCRTOC. Without changing existingcamera placements or adding extra cameras, point cloud frame mining can serveas a free lunch to improve 3D manipulation learning.

Frame Mining: a Free Lunch for Learning Robotic Manipulation from 3D Point Clouds