NVIDIA Launches Newton 1.0 for GPU-Accelerated Physics in Industrial Robot Training
NVIDIA launched Newton 1.0 at GTC 2026, a physics simulation engine designed for industrial robot training, offering a 475x speed increase over Google DeepMind's MJX on RTX PRO 6000 Blackwell GPUs. The open-source engine focuses on contact-rich simulations for industrial assembly.
The robotic simulation market is nearing $28 billion, with manufacturers striving to improve sim-to-real transfer. Key features include advanced collision detection and hydroelastic modeling. Partnerships involve Skild AI for reinforcement learning in assembly tasks and Samsung for synthetic training data generation. Newton is housed under the Linux Foundation, in collaboration with Google DeepMind and Disney Research, and it ships under open-source licensing.
