NVIDIA has released Newton 1.0, a production-ready open-source physics simulation engine for training robots on contact-rich manipulation and locomotion tasks in industrial settings.
The release marks the transition from beta, first unveiled in September 2025, to a stable platform already being adopted by manufacturers for real-world assembly automation. Newton 1.0 is built on NVIDIA Warp and OpenUSD and co-developed with Google DeepMind and Disney Research under the Linux Foundation.
The engine ships with multiple rigid-body solvers, including MuJoCo Warp and Kamino. According to NVIDIA’s technical blog, MuJoCo Warp running on the NVIDIA RTX PRO 6000 Blackwell GPU achieves up to 252 times the speed of MJX for locomotion tasks and up to 475 times faster for manipulation tasks. The engine also includes a Vertex Block Descent solver for deformable simulation of cables, cloth, and volumetric materials, alongside a signed-distance-field collision library and hydroelastic contact modeling.
Newton integrates natively with Isaac Lab 3.0, NVIDIA’s open-source robot learning framework, which was also released in early access on Monday. Toyota Research Institute has partnered with NVIDIA to advance solver development and contact modeling.
Skild AI is using Newton with Isaac Lab to train reinforcement learning policies for GPU rack assembly, focusing on connector insertion, board placement, and fastening with tight tolerances. Samsung is collaborating with Lightwheel to utilize Newton’s deformable simulation for cable manipulation in refrigerator assembly lines.
NVIDIA unveiled additional robotics announcements at GTC, including Cosmos 3, GR00T N1.7, and a preview of GR00T N2. Jensen Huang said GR00T N2 helps robots succeed at new tasks in new environments more than twice as often as leading alternatives. The company said its Isaac and Omniverse frameworks now reach an installed base exceeding 2 million robots globally.
