From Human Hands to Robot Hands
From large-scale human hand data collection to robot dexterous manipulation — A research strategy survey
Human hand data and tactile sensing to train robot manipulation policies. 4 Parts, 11 Chapters.
First published: 2026-04-07 | Last updated: 2026-04-07
Why Human Hand Data
Going beyond the structural limitations of teleoperation to teach robots with large-scale data from natural human activities.
What This Book Offers
An expert-level research strategy survey. Focuses on specific metrics, limitations, research group analysis, and novelty definition.
TacGlove + TacTeleOp + TacPlay
A rationale document for Prof. Park Y.-L.'s tactile data glove (TacGlove), Prof. Park J.-W.'s multi-object grasping co-training (TacTeleOp), and autonomous play learning (TacPlay) research.
Part I: Background and Motivation
Why Human Hand Data? — Limitations of Teleoperation and Alternatives
Data bottleneck, three limitations of teleoperation, teleop data vs human hand data
→ 02Collecting Human Hand Data — From Sensors to Datasets
Motion gloves, tactile gloves, exoskeletons, egocentric vision, dataset comparison
→ 03The Value of Touch — Why Force Information Is Essential
Contact-rich tasks, tactile sensor technology, superiority of tactile-informed policies
→Part II: Human Data to Robot Policy
Can Human Data Alone Suffice? — Teleop-Free Approaches
AirExo, DEXOP, EgoScale, AoE and other teleop-free learning cases
→ 05Synergy of Teleoperation and Human Data — Co-training
Cross-embodiment data mixing, co-training strategies, scaling laws
→ 06Bridging the Embodiment Gap — Retargeting and Alignment
Kinematic/visual/tactile retargeting, latent space alignment
→Part III: Our Proposals
TacGlove — Stretchable Tactile Data Glove
Stretchable glove, simultaneous kinematics + tactile sensing, data glove survey
→ 08TacTeleOp — Multi-Object Grasping and Tactile Co-training
ROBOTIS HX5-D20, multi-object grasping, teleop + human data co-training
→ 09TacPlay — Teleop-Free Autonomous Learning
Tactile play, autonomous embodiment gap learning, complete teleop elimination
→