stable-envs-v1: open-source GPU-native rendering speeds world model training up to 1.33x across 80 environments
tensorqt · x · 2026-11-19
Researchers released stable-envs-v1 (arXiv:2611.23151), an open-source library of 80 simulation environments with GPU-native, on-demand rendering for world model training.
- Core idea: WM training usually shuttles large image tensors between memory and compute; stable-envs renders observations directly on GPU, boosting utilization.
- Results: end-to-end training speedups up to 1.33x with no performance loss, plus batched physics for reproducible large-scale evaluations with full determinism.
- Controllable variations: visual changes applied on-the-fly at render time increase training diversity.
- Supports multiple CPU/GPU backends; code is openly available.
More from Embodied
- BrightAI raises $97M to replace Roman-style infrastructure inspections with physical AI — pzakin · 2026-11-19
- π0.7 drives a real Tesla with 1.5 hours of data; Scoble urges Tesla to open source everything — Scobleizer · 2026-11-19
- Robotics Team Offers $5K Fixed-Scope Sprints to Clear Hardware Bottlenecks — Scobleizer · 2026-11-19
- ETH's Anymal robot completes 10km autonomous alpine hike with VLM-guided ROAM system — ChongZzZhang · 2026-11-19
- iFixit tears down Apple HomeView: lots of glue and 'endless frustration' — Resident-Fall4351 · 2026-11-19
- Removing human video drops robot OOD success from 60% to 7%: generalization comes from human video — gui_penedo · 2026-11-19