Open-source Unitree G1 motion imitation extends GfR and HIL to two tasks

ChongZzZhang · x · 2026-07-28

Unofficial Hybrid Motion Imitation for Unitree G1

• The post points to an unofficial open-source implementation and extension of two papers, GfR and HIL, built on top of Holosoma.
• It targets the 29-DoF Unitree G1 and includes two tasks: box moving and box climbing.
• The key idea is that the same hybrid learning framework can handle both locomotion and loco-manipulation.
• The author says the method can work with only one reference motion clip per task, and the learned policies generalize across a wide range of conditions and goals.
• A follow-up comment notes the approach is easy to implement and gives decent results, but may overfit to random distribution bias; still, it provides useful inspiration.

Original post →

More from Embodied

• Local LLM agent drives Perfect Dark on Mac with 50 ms combat control — moyoteg · 2026-07-28
• NoriRobotics Robots Roll Off the Production Line — chris_j_paxton · 2026-07-28
• Safe Driving Achieved with Only Cameras and AI, Major Implications for Autonomy Scaling — MickeySteamboat · 2026-07-28
• Salamander-inspired CPG demo switches from walking to swimming with one input — zzznah · 2026-07-28
• manosim: Open-Source High-Precision MANO Hand Models for MuJoCo — andrew_n_carr · 2026-07-28
• US Scientists Create 99.9999% Ultra-Pure Silicon for Quantum Chips, Eliminating Atomic Noise — Dr_Singularity · 2026-07-28

Embodied channel →