HumanPlus

HumanPlus is a comprehensive humanoid robot learning system developed by researchers at Stanford University. The system provides a full pipeline for whole-body humanoid teleoperation and imitation learning, enabling a Unitree H1 humanoid robot to shadow human motion in real time and learn skills from single-camera human video data. The dataset includes 40 real-world demonstration tasks spanning both locomotion (walking, turning) and manipulation (picking, placing, pushing), collected through a wearable teleoperation suit. HumanPlus demonstrated that humanoid robots can learn complex whole-body skills by mimicking human motion, achieving high success rates on both seen and unseen tasks. The project is fully open-sourced with MIT license, and has become one of the most cited humanoid learning platforms since its release in 2024.