Weshow that a single generalist policy can perform in-hand manipulation of over100 geometrically-diverse real-world objects . Multi-task learning withobject point cloud representations not only generalizes better but evenoutperforms the single-object specialist policies on both training as well as held-out test objects . Video results athttps://huangwl18.io/geometry-dex.com/video results at http://www.hug.org/weshow-how-wescohen/video-video-study-to-the-robot-learners .

Author(s) : Wenlong Huang, Igor Mordatch, Pieter Abbeel, Deepak Pathak

Links : PDF - Abstract

Code :

Keywords : video - multi - results - weshow - geometry -

Leave a Reply

Your email address will not be published. Required fields are marked *