← Back
NVIDIA
NVIDIA releases Newton 1.0, GPU-accelerated physics engine for industrial robotics with 252x MuJoCo speedup
· releasefeatureplatformperformanceopen-source · developer.nvidia.com ↗

Production-Ready Physics Simulation for Robotics

NVIDIA announced Newton 1.0 GA at GTC 2026, a GPU-accelerated physics engine designed to balance speed and realism for industrial robotics applications. Built on NVIDIA Warp and OpenUSD, Newton provides a production-ready foundation for dexterous manipulation and locomotion tasks, with integrations into NVIDIA Isaac Lab and Isaac Sim.

Modular Architecture and Multi-Solver Approach

Newton's key strength is its modular design that supports multiple solvers and simulation components through a unified interface. Rather than forcing users into a single physics engine, it accommodates common robotics description formats (MJCF, URDF, OpenUSD) and allows teams to mix and match collision detection, contact models, sensors, and solver backends while maintaining a consistent simulation stack.

Advanced Solvers and Significant Performance Gains

The framework ships with versatile rigid-body solvers:

  • Kamino (Disney Research): Handles complex mechanisms like robotic hands and legged systems with closed-loop linkages and passive actuation, enabling mechanical designers greater freedom without simulatability constraints
  • MuJoCo 3.5 (MJWarp) (Google DeepMind): GPU-scaled version achieving 252x speedup for locomotion and 475x for manipulation tasks on NVIDIA RTX PRO 6000 Blackwell, extending the stability roboticists trust in MuJoCo with massive parallelism

Rich Simulation Capabilities for Real-World Scenarios

Newton addresses real industrial needs with advanced features:

  • Deformable simulation: Vertex Block Descent (VBD) solver handles cables, cloth, and rubber parts; Implicit Material Point Method (iMPM) supports granular materials for rough terrain
  • Hydroelastic contacts: High-fidelity pressure distributions across contact patches enable better tactile sensing and manipulation policies with improved sim-to-real transfer
  • SDF-based collision: Direct CAD mesh processing eliminates approximation needs for precision tasks like connector insertion
  • Tiled camera sensor: Warp-based rendering supports RGB, depth, albedo, normals, and instance segmentation for high-throughput perception data

Integration with Isaac Ecosystem

Native OpenUSD integration with NVIDIA Isaac Sim 6.0 and Isaac Lab 3.0 accelerates workflows from robot description through policy training and evaluation, supporting both reinforcement and imitation learning pipelines.