Cartwheel open-sources three humanoid motor control policies, G1 lands a cartwheel
andrew_n_carr · x · 2026-11-11
Cartwheel founder Andrew Carr introduced Cartwheel Control, a hardware-agnostic approach for training whole-body motor control policies from large-scale kinematic motion, alongside three open motion-tracking policies runnable in 2 minutes.
- Covers three morphologies: Unitree G1 (29 joints), Menlo Asimov-1 (25 joints), and RoboParty RP1 (28 joints)
- Policies drive joints at 50 Hz to track reference motions from subtle weight shifts and footwork to high kicks and ground recoveries—G1 performs a real cartwheel, RP1 sticks a 360-degree backflip into a kneeling landing
- Weights on Hugging Face, Apache-2.0 simulation/visualization code on GitHub, built on the open ecosystem (BeyondMimic, MuJoCo Warp, mjlab)
Carr argues kinematics should not be treated as legacy scaffolding: generative kinematics alone is a physics-free ghost, motor control alone has balance but no semantics. Stacking generative kinematics on a solid 50 Hz tracking policy means every advance in motion generation immediately upgrades the robot—no thousands of hours of fresh teleop needed.
More from Embodied
- ETH researcher vibe-codes SuperMimic humanoid reproduction in 1h, LAFAN passes in 1.5 GPU-hours — ChongZzZhang · 2026-11-11
- Open-source companion robot Lamp ships at $699 with custom personality, skills and an app economy — dee_hw · 2026-11-11
- Must-read papers of the week: HALO humanoid motion, GR00T 2, VLA policies, agent tooling — TheTuringPost · 2026-11-11
- Microsoft's DLSS rival Auto SR launches on Xbox Ally X with handheld 1080p and 120fps support — tomwarren · 2026-11-11
- NVIDIA's PhysicalAI-Human-1M-100K robot dataset trends on Hugging Face — nvidia · 2026-11-11
- Developer runs a hermes-agent lineup on ESP32 with voice-to-voice and tool calling — alexcovo_eth · 2026-11-11