Skip to content

MOSAIC (humanoid motion tracking and teleoperation)

  • AI model
  • Humanoids and robot learning

MOSAIC is a training codebase for humanoid motion tracking and whole-body teleoperation, built on Isaac Lab and aimed at G1, H1_2 and Adam robots.

Made by BAAI-Humanoid, China (checked on github.com/BAAI-Humanoid (opens in a new tab)).

Source on GitHub: github.com/BAAI-Humanoid/MOSAIC (opens in a new tab). Facts read at commit d73e885 (opens in a new tab).

What it does

  • Trains a teleoperation-oriented motion tracker that emphasizes global motion consistency for long-horizon behaviors.
  • Adds interface-specific corrections through rapid residual adaptation to bridge sim-to-real gaps.
  • Converts retargeted motion CSV files into full kinematic data via forward kinematics for training.
  • Provides task IDs for single-motion and multi-motion tracking, distillation, and multi-teacher residual training.
  • Supports multi-GPU training through torchrun and logs runs to WandB.
  • Deployment to simulation or a real robot is handled by a separate project, RobotBridge.

Written by Robotheca from the project's README (opens in a new tab) at commit d73e885.

Getting started

Difficulty
Advanced

Maintenance

Status
Maintained: a commit in the last 12 months
Newest commit
7 Mar 2026 on main
Latest release
None published on GitHub

Checked on GitHub on 16 Oct 2026.

Robots that can run it

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.