Podcast Episode
Co-developed with Google DeepMind and Disney Research under the Linux Foundation, Newton 1.0 is built on NVIDIA Warp and OpenUSD. It ships with multiple rigid-body solvers, including MuJoCo Warp, a GPU-optimised port of Google DeepMind's widely used MuJoCo simulator, and Kamino, a Disney Research solver built for complex closed-chain mechanisms like robotic hands and legged systems.
NVIDIA Launches Newton 1.0: Open-Source Physics Engine Built to Train Robots
March 17, 2026
0:00
3:00
NVIDIA has released Newton 1.0 at GTC 2026, an open-source GPU-accelerated physics engine co-developed with Google DeepMind and Disney Research for training robots on complex real-world tasks. The engine delivers simulation speeds hundreds of times faster than existing tools and is already being used by manufacturers for factory automation.
A New Engine for Robot Intelligence
NVIDIA has officially launched Newton 1.0, a production-ready, open-source physics simulation engine designed specifically for training robots on contact-rich manipulation and locomotion tasks. Announced at GTC 2026 in San Jose, the release marks Newton's transition from beta to a stable platform already seeing real-world adoption.Co-developed with Google DeepMind and Disney Research under the Linux Foundation, Newton 1.0 is built on NVIDIA Warp and OpenUSD. It ships with multiple rigid-body solvers, including MuJoCo Warp, a GPU-optimised port of Google DeepMind's widely used MuJoCo simulator, and Kamino, a Disney Research solver built for complex closed-chain mechanisms like robotic hands and legged systems.
Massive Speed Gains
Performance benchmarks are striking. Running on NVIDIA's RTX PRO 6000 Blackwell GPU, MuJoCo Warp achieves up to two hundred and fifty-two times the speed of MJX for locomotion tasks and up to four hundred and seventy-five times faster for manipulation tasks. The engine also supports deformable simulation through the Vertex Block Descent solver, handling cables, cloth, and volumetric materials with high fidelity.From Simulation to the Factory Floor
Early adopters are already putting Newton to work. Skild AI is using Newton with Isaac Lab to train reinforcement learning policies for GPU rack assembly, a demanding electronics manufacturing task requiring connector insertion, board placement, and fastening with tight tolerances. Samsung, working with Lightwheel, is applying Newton's deformable simulation to cable manipulation on refrigerator assembly lines.Part of a Broader Robotics Push
Newton 1.0 arrives alongside several major announcements from NVIDIA, including Cosmos 3, a world foundation model for robot intelligence; GR00T N1.7, a commercially licensed humanoid robot model; and a preview of GR00T N2, which reportedly helps robots succeed at new tasks more than twice as often as leading alternatives. Toyota Research Institute has also joined as a partner to advance solver development. NVIDIA says its Isaac and Omniverse frameworks now reach an installed base exceeding two million robots globally.Published March 17, 2026 at 2:17am