Speakers
Description
How can humanoid robots learn complex manipulation skills from human demonstrations? Teleoperation provides a practical way to transfer human movements to robots while simultaneously collecting data for training intelligent robotic systems.
In this presentation, we demonstrate teleoperation of the Unitree G1 humanoid robot using OptiTrack motion capture and the PICO 4 Ultra VR headset, combined with MANUS Meta Gloves Pro Haptic for finger tracking and dexterous hand control.
We explain how human body and hand movements are mapped to the robot's joints and how synchronized data from motion capture, cameras, joint sensors, and tactile sensors can be collected to create datasets for robot learning.
We discuss how these demonstrations can support the development of Vision-Language-Action (VLA) models and World Action Models (WAM),
enabling robots to learn, predict, and eventually perform manipulation tasks autonomously.
The presentation highlights practical challenges in human-to-robot motion transfer, data collection, and learning, accompanied by a demonstration of humanoid teleoperation.