← Back
NVIDIA
NVIDIA Newton 1.0 GA ships GPU-accelerated physics simulator with 252x-475x speedups for robot learning
· releaseplatformfeatureperformanceopen-sourcesdk · developer.nvidia.com ↗

Newton 1.0 GA: Production-Ready Robotics Physics Engine

NVIDIA announced Newton 1.0 GA at GTC 2026, marking the first production-ready release of its GPU-accelerated physics simulator specifically engineered for complex robotic tasks. Newton combines high-speed simulation with realistic physics modeling, addressing the traditional tradeoff between performance and accuracy that has constrained robotics research and development.

Key Performance Gains

The release delivers substantial performance improvements, with MuJoCo Warp achieving 252x speedup for locomotion tasks and 475x speedup for manipulation on NVIDIA RTX PRO 6000 Blackwell Series GPUs. These optimizations enable training thousands of parallel robotic environments, dramatically accelerating reinforcement learning and imitation learning workflows.

Modular Architecture and Solver Options

Newton's strength lies in its modular design, supporting multiple solver backends and collision detection approaches:

  • MuJoCo Warp (MuJoCo 3.5): GPU-accelerated version of the trusted MuJoCo simulator, now with massive parallel throughput
  • Kamino: Disney Research's advanced solver for complex mechanisms, robotic hands, and legged systems with closed-loop linkages and passive actuation
  • Deformable Solvers: Vertex Block Descent (VBD) for cables, cloth, and rubber parts; Implicit Material Point Method (iMPM) for granular materials and rough terrain
  • Flexible Collision Detection: SDF-based collision for CAD-exported meshes and hydroelastic contacts for high-fidelity tactile simulation

OpenUSD Integration and Developer Workflow

The simulator natively integrates with NVIDIA Isaac Sim 6.0 and Isaac Lab 3.0, using OpenUSD as a common data layer. This enables seamless workflows from robot description through policy training and evaluation. Support for standard robotics formats (MJCF, URDF, OpenUSD) allows teams to leverage existing robot assets without reimplementation.

Availability and Open Source

Newton is available as an open-source project, making it accessible to the robotics community while providing a stable API for modeling, solving, controlling, and sensing capabilities in simulation.