Genesis World 1.0
A tiernew this weekOpen-source robotics simulation stack that compresses 200+ hours of real-world policy evaluation down to ~30 minutes using GPU-parallelized physics and photorealistic rendering.
Kai's verdict
If you're building or evaluating robotics foundation models, this is the most credible open-source sim stack to emerge in years — the real-to-sim correlation numbers are genuinely impressive, and the full-stack approach (physics + renderer + compiler) is more coherent than duct-taping MuJoCo to a game engine. That said, it's a 1.0 from a company validating its own benchmarks, so trust but verify. (Verdict pending Phi's full review.)
Strengths
- Collapses multi-day real-world robot eval to ~30 minutes via massively parallel GPU simulation
- 89% Pearson correlation with on-hardware rollouts — claims a 45% smaller sim-to-real gap than alternatives
- Unified multi-physics engine (rigid bodies, FEM, MPM, SPH, PBD) behind a single Pythonic API
- Nyx renderer hits noise-free 1080p in 4ms or less, designed specifically for robot camera fidelity — not repurposed game engine tricks
- Quadrants compiler targets CUDA, ROCm, Metal, Vulkan, x86, and ARM64 from the same Python kernel code
Weaknesses
- Heavily GPU-dependent — Nyx requires an NVIDIA GPU with CUDA; not a casual install for the average developer
- Squarely aimed at robotics researchers and ML engineers building foundation models — no web UI, no no-code path
- Brand-new 1.0 release with limited community validation outside Genesis AI's own benchmarks
Best for
Robotics ML researchers and engineers who need to run large-scale policy evaluation and ablations without burning hundreds of hours on physical hardware.
Pricing
Free (Apache 2.0 open source)
Fully open source. Nyx renderer plugin (gs-nyx-plugin) also free via pip. Requires your own NVIDIA GPU hardware for full performance.