Humanoid Manipulation Interface: Humanoid Whole-Body Manipulation from Robot-Free Demonstrations

[Project Page] [Paper] [Dataset] [Hardware]

Ruiqian Nai*, Boyuan Zheng*, Junming Zhao*, Haodong Zhu, Sicong Dai, Zunhao Chen, Yihang Hu, Yingdong Hu, Tong Zhang, Chuan Wen, Yang Gao

* Equal contribution

This repository contains the full stack for HuMI (Humanoid Manipulation Interface), including data collection, policy training, and deployment. Each component is organized into a separate folder.

👀 Overview

.
├── humi_data_collection
├── humi_deploy_high_level
├── humi_deploy_low_level
├── humi_high_level_policy
└── humi_low_level_controller

• humi_data_collection: Code for collecting and processing humanoid-free demonstrations.
• humi_high_level_policy: Code for training the high-level policy (diffusion policy).
• humi_low_level_controller: Code for training the low-level controller.
• humi_deploy_high_level and humi_deploy_low_level: Code for deploying the trained policies on a humanoid robot.

🔧 Installation

See the README.md file in each folder for detailed installation instructions.

📘 Instructions

Topic	Instructions
Hardware instruction	hardware_guide.md
Data collection	humi_data_collection/README.md
High-level policy training	humi_high_level_policy/README.md
Low-level controller training	humi_low_level_controller/README.md
Policy deployment	deploy_instruction.md