Robotic Telekinesis: Learning a Robotic Hand Imitator by Watching Humans on YouTube

Abstract

We build a system that enables any human to control a robot hand and arm, simply by demonstrating motions with their own hand. The robot observes the human operator via a single RGB camera and imitates their actions in real-time. Human hands and robot hands differ in shape, size, and joint structure, and performing this translation from a single uncalibrated camera is a highly underconstrained problem. Moreover, the retargeted trajectories must effectively execute tasks on a physical robot, which requires them to be temporally smooth and free of selfcollisions. Our key insight is that while paired human-robot correspondence data is expensive to collect, the internet contains a massive corpus of rich and diverse human hand videos. We leverage this data to train a system that understands human hands and retargets a human video stream into a robot hand-arm trajectory that is smooth, swift, safe, and semantically similar to the guiding demonstration. We demonstrate that it enables previously untrained people to teleoperate a robot on various dexterous manipulation tasks. Our low-cost, glove-free, markerfree remote teleoperation system makes robot teaching more accessible and we hope that it can aid robots in learning to act autonomously in the real world. Video demos can be found at: https://robotic-telekinesis.github.io

Learning dexterity from human hand motion in internet videos

“…We call our system Robotic Telekinesis as it provides a human the ability to control a dexterous robot from a distance without any physical interaction as in Figure 2. This builds on work published at RSS 2022 by Sivakumar et al, (2022).…”

“…Our key insight is to leverage the vast corpus of human hand poses from passive data on the web to train a retargeting system from human pose to robot pose as seen in Figure 1. This neural network learns to map across the large embodiment gap between human and robot as described in Handa et al (2020) and Sivakumar et al (2022). It uses unpaired data of humans using their hands, and uses this to learn how to control a robot hand.…”

“…We call our system Robotic Telekinesis as it provides a human the ability to control a dexterous robot from a distance without any physical interaction as in Figure 2. This builds on work published at RSS 2022 by Sivakumar et al, (2022).

Figure 2.An operator completing a dice pickup task while watching the robot through a video conference. Video demos are at https://robotic-telekinesis.github.io/.

…”

LEAP Hand: Low-Cost, Efficient, and Anthropomorphic Hand for Robot Learning

“…In contrast to tendon-driven hands, which have motors in the wrist, the Allegro Hand [20] has its motors in the finger joints. It is most popular in research labs [31,32,33,34,35] because it is relatively cheap (16k USD). However, users find the motors in the fingers to be weak for many everyday tasks.…”

“…from MANO [52] parameters which parameterize a human hand. Closely related is the teleoperation of robot hands from real-time video [31,53], which can be used to guide learning and improve sample-efficiency [54,55]. Hand poses can be extracted from video data available on the web to learn manipulation policies [54,56].…”