SG-JEPA Paper: World Models That Train on Earth and Deploy on Mars, Halving Zero-Shot Physics Error
randall_balestr · x · 2026-09-11
A new arXiv paper (2609.10464) introduces Semigroup-JEPA (SG-JEPA), testing whether JEPA world models can zero-shot generalize to dynamics never seen in training — "train on Earth, deploy on Mars." It conditions the temporal model on physics parameters via action-conditioning and jointly trains encoder and predictor with autoregressive latent rollout. Evaluated on tasks across gravitational fields, SG-JEPA cuts open-loop prediction error up to 2x vs DINO-WM on 2D data and boosts control success up to 2.5x on 3D robotic datasets with diffusion policies. A linear feature model explains the gains: backpropagating multi-step rollout loss into representations trains the encoder to preserve physics-consistent features.
More from Embodied
- Claude Code finds root backdoor in Honda Civic head unit, then debloats it — IndraVahan · 2026-10-26
- Temple health wearable opens pre-orders, users ask: why wear it before benefits are proven — abhishekcode42 · 2026-10-26
- FLASH: GPU-native deformable sim learns cloth folding in minutes on one RTX 5090 — ChongZzZhang · 2026-10-26
- Luke Wroblewski spots it: iPhone Duo's corner radius mirrors the original iPhone — LukeW · 2026-10-26
- Humanoids shattered forecasts: half-marathon record in 2026, 100m faster than Bolt — ben_j_todd · 2026-10-26
- Humanoid walking is continuous correction — watch recovery, not locomotion, in demos — Olivier__OG · 2026-10-26