March 19, 2026InfrastructureOpen SourceAgents

Newton 1.0: Open-Source Physics Engine for Training Robotics Agents

Newton 1.0, the open-source GPU-accelerated physics simulation engine, has reached general availability. Announced at NVIDIA GTC 2026 on March 17, it provides a production-ready foundation for training dexterous manipulation and locomotion agents in simulation.

Newton was initiated by Disney Research, Google DeepMind, and NVIDIA, and is now a Linux Foundation project. Built on NVIDIA Warp with MuJoCo Warp as its primary backend, Newton emphasizes GPU-based computation, OpenUSD support, differentiability, and user-defined extensibility.

The performance numbers are striking: MuJoCo Warp running on the NVIDIA RTX PRO 6000 Blackwell GPU achieves up to 252x the speed of MJX for locomotion tasks and up to 475x faster for manipulation tasks. This level of speedup means robotics researchers can run millions of simulation episodes in hours instead of weeks.

Newton is directly relevant to the agentic ecosystem as embodied AI agents — robots that interact with the physical world — need fast, accurate simulation for training. The engine supports the full pipeline from rapid prototyping to scalable training of robot agents.

GitHub: https://github.com/newton-physics/newton
NVIDIA Developer: https://developer.nvidia.com/newton-physics
← Previous
Google DeepMind Releases AGI Cognitive Framework with $200K Kaggle Hackathon
Next →
Online Experiential Learning: Microsoft's Framework for Agents That Improve from Deployment
← Back to all articles
>_