Start here. Tiny task-free world models, frozen, planned live: quads, a plane, a cable robot, arms. The front door, papers, models.
🦆 locked-in brother.
Edgar Moreau
Ethgar
AI & ML interests
Autonomous UAVs, Embodied AI, World models, application over research.
Recent Activity
liked a model about 24 hours ago
wkentaro/sam3-onnx-models-v0.5.0 liked a Space 1 day ago
lerobot/language-conditioned-policies upvoted a paper 1 day ago
Energy-Based Transformers are Scalable Learners and ThinkersOrganizations
G1 policies trained in mjlab on HF Jobs (~$1 each), measured for energy: walk/stand, carry from PRISM clips.
June 2026, before world models: rescue flights over real Alps terrain and tiny MLP policies. Rescue is now a world-model mission.
Start here. Tiny task-free world models, frozen, planned live: quads, a plane, a cable robot, arms. The front door, papers, models.
Clean-room SkyJEPA reimplementation: a 5.6k-parameter JEPA world model flown by MPPI, its data, and the v2 that was not promoted.
G1 policies trained in mjlab on HF Jobs (~$1 each), measured for energy: walk/stand, carry from PRISM clips.
Before world models: teleop data, the first policy run on the real arm (ACT, Aug 2026), ACT and SmolVLA in a desk twin.
June 2026, before world models: rescue flights over real Alps terrain and tiny MLP policies. Rescue is now a world-model mission.