Unitree open-sources 6B humanoid model UnifoLM-WLA-1.0, claims fine-tuning on a 24GB GPU
lmoroney · x · 2026-10-12
Unitree released UnifoLM-WLA-1.0, a 6B-parameter model driving its G1 humanoid through 64 tasks (54 tabletop, 10 whole-body), with Apache 2.0 weights (12.5GB checkpoint on Hugging Face).
The default fine-tuning recipe freezes the vision-language backbone and trains only the action head and robot-state projector, with LoRA optional. Caveats: the server currently only supports the two-finger-gripper action format, and no per-task success rates have been published. Practical tip: use the bundled script to replay a recorded episode and plot predicted vs. recorded joint actions before touching real hardware.
Related event: Unitree Open-Sources 6B Humanoid Robot Model UnifoLM-WLA-1.0(2 posts)→
More from Embodied
- Running Two AI Agents on Raspberry Pis with Full Sudo, RTL-SDRs and Starlink — natesiggard · 2026-10-12
- Tesla Cybercab tracker logs 145 on the road, 319 registered with Texas DMV — JOBhakdi · 2026-10-12
- Liquid AI open-sources Open d1 decision models: d1-3B runs at 35ms per image on Jetson Thor — TheTuringPost · 2026-10-12
- Fleet Week traffic trapped Andrew Chen's Waymo for 15 min until humans-in-the-loop rerouted it — andrewchen · 2026-10-12
- COMODO distills video intelligence into IMU sensors for camera-free egocentric AI glasses — flosalim · 2026-10-12
- DirtyMoCap recovers 3D motion from noisy unordered markers, with 100x faster CUDA solver — jamestagg · 2026-10-12