A System for General In-Hand Object Re-Orientation

Tao Chen     Jie Xu     Pulkit Agrawal
Computer Science and Artificial Intelligence Laboratory (CSAIL)
Massachusetts Institute of Technology


Abstract


In-hand object reorientation has been a challenging problem in robotics due to high dimensional actuation space and the frequent change in contact state between the fingers and the objects. We present a simple model-free framework that can learn to reorient objects with both the hand facing upwards and downwards. We demonstrate the capability of reorienting over 2000 geometrically different objects in both cases. The learned policies show strong zero-shot transfer performance on new objects. We provide evidence that these policies are amenable to real-world operation by distilling them to use observations easily available in the real world.


Paper


A System for General In-Hand Object Re-Orientation
Tao Chen, Jie Xu, Pulkit Agrawal
Conference on Robot Learning (CoRL), 2021 (Best Paper Award) (Oral , acceptance rate: 6.5%)
paper / arXiv / project page / bibtex

Press coverage: MIT News, MIT CSAIL News, AZO Robotics, AIHub, AI科技评论, Tech Xplore, Communications of the ACM, Inceptive Mind, IEEE Spectrum, The Hack Posts, Tectales, The Robot Report


Oral Talk





Teaser





Video Demo





More Videos


    Hand faces downward

    (with a table)

    More videos here

    Hand faces downward

    (in the air)

    More videos here

    Hand faces upward
    More videos here

    Lifting
    More videos here

    Vision policy
    More videos here