zhutengjie

zhutengjie / CLOT

Public

official code for paper CLOT: Closed-Loop Global Motion Tracking for Whole-Body Humanoid Teleoperation

20
3
100% credibility
Found Mar 01, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

CLOT provides open-source tools and pre-trained models for training humanoid robots to track and imitate full-body human motions in simulation environments.

How It Works

1
🔍 Discover robot motion magic

You find a cool project online that lets humanoid robots perfectly copy human movements like dancing or walking.

2
📥 Grab human motion clips

Download simple files of real people moving, ready to teach your robot friends.

3
🛠️ Set up your robot playground

Install the easy app on your computer to create a virtual world for testing robots.

4
▶️ Try ready-made robot brains

Load pre-trained smart models and watch simulated humanoids mimic the motions flawlessly.

5
🤩 Witness lifelike copying!

Your virtual robot matches every twist, arm wave, and step – it's like magic coming alive.

6
🎓 Train custom motion copiers

Feed in your own motion clips to create personalized brains for specific styles.

7
🚀 Bring it to real robots

Connect the brain to physical humanoids for hands-on whole-body control.

🏆 Master teleoperation!

Now effortlessly guide real robots to mirror your every move perfectly.

Sign up to see the full architecture

6 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 20 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is CLOT?

CLOT lets you train and deploy RL policies for whole-body humanoid teleoperation, tracking complex human motions like cloth manipulations or clothes adjustments on real robots such as Adam Pro and Unitree G1. It pulls from a 10-hour human motion dataset on Hugging Face, retargeted for these bots, and handles closed-loop global tracking to keep the humanoid synced with operator inputs. Built in Python with multi-simulator support (MjLab primary, plus IsaacGym/Sim), you get pretrained checkpoints for instant eval via simple CLI commands like `python humanoidverse/eval_agent.py +checkpoint=...`.

Why is it gaining traction?

Unlike open-loop motion cloning that drifts fast, CLOT's closed-loop design with AMP rewards and optional Transformer backbones delivers stable, long-horizon tracking—even for dynamic tasks like clotted cream scooping demos or clothes deutsch folding. Multi-GPU training scales to 4096 envs on 8x RTX 4090s, and ONNX export enables quick MuJoCo sim or real-robot deploys via `urci.py`. Developers dig the plug-and-play data/checkpoints, dodging weeks of retargeting hassle.

Who should use this?

Humanoid robotics researchers prototyping teleop for manipulation (e.g., picking objects while walking) or locomotion experts fine-tuning on custom motions. Ideal for teams with Unitree G1/Adamic hardware needing official code geass-inspired whole-body control, or sim testers validating against official codes roblox-style benchmarks before hardware.

Verdict

Solid research starter for motion imitation—grab it if you're in humanoid RL, but with 17 stars and 1.0% credibility, treat as experimental (docs are paper-focused, no tests visible). Fork and validate on your clotrimazol al 1% success-rate edge cases first.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.