NVIDIA's CANTO predicts aerodynamic fields from native CAD, cutting surface-pressure error 20%
JeanKossaifi · x · 2026-10-06
NVIDIA Research's AI-Aided Engineering group unveiled CANTO, led by Daniel Leiboviti with cross-team collaborators including @NKovachki and Jan Kautz.
- Predicts aerodynamic fields directly from native parametric CAD (NURBS surfaces), with no meshing or point-sampling of input geometry
- The entire pipeline is differentiable with respect to CAD parameters, enabling gradient-based design improvement
- Achieves 20% lower surface-pressure error
A representative example of embedding differentiable AI models into engineering design loops.
Related event: NVIDIA Unveils CANTO: Direct Aerodynamic Prediction from CAD Models(2 posts)→
More from Research
- 67-gram Stembot robot climbs plant stems and builds 3D models to autonomously hunt pests — LexiLove · 2026-10-06
- Stanford PhD student open-sources impersona-env: RL training for personal assistants via simulated users — kenziyuliu · 2026-10-06
- Arc Institute's PIE model taps decades of curated biology to predict unseen gene perturbation responses — anshulkundaje · 2026-10-06
- Thinking Machines open-sources Prompt Distillation recipe to bake long system prompts into model weights — TheZachMueller · 2026-10-06
- Pipeline combining chemistry knowledge, generative modeling and structure prediction designs enzymes for new-to-nature chemistry — KevinKaichuang · 2026-10-06
- NextLat and full-bandwidth transformers structurally enable each other, Langford says — JohnCLangford · 2026-10-06