Nvidia teams up with Disney and DeepMind on open-source physics engine Newton

Newton is an upcoming open-source physics engine developed jointly by Nvidia, Google DeepMind, and Disney Research. Its main job is simple: give robots a high-fidelity physical sandbox where they can train and master complex, nuanced movements with far greater precision than before.

What Newton actually does

Most modern robot development hinges on simulation, but bridging the sim-to-real gap is notoriously hard. When physics in the simulator aren't accurate enough, trained behaviors fail the moment a machine touches real-world ground. Newton aims to tighten that accuracy gap specifically for machine learning workflows.

Disney Research will be one of the very first teams to deploy Newton in production. Disney plans to plug the engine into its robotic character platform to power next-generation entertainment animatronics. A prime example is the expressive, Star Wars-inspired BDX droid that paced across the stage with Jensen Huang at GTC. Instead of relying purely on rigid, pre-programmed motion paths, robots powered by Newton can learn to walk, interact, and emote in real time.

How it works and availability

Building expressive physical movement requires blending heavy physics simulation with deep reinforcement learning. Newton pulls together Nvidia's computing infrastructure, Google DeepMind's robotics learning research, and Disney's storied expertise in character animation.

According to Disney Imagineering R&D, this partnership will allow them to bring lifelike characters to life in ways the world hasn't seen before, making theme park interactions far more dynamic.

As announced at GTC, Newton is slated to be fully open-source. The project is currently in active joint development across the three organizations, with codebase releases and developer access details expected as the platform matures.

Why it matters

If you work in embodied AI, quadruped control, or robotics research, Newton is an immediate project to keep on your radar. Many existing simulators force you to choose between game-engine speed and scientific-grade physical accuracy.

Bringing DeepMind's algorithmic chops and Disney's master-level kinematics directly into an open-source Nvidia engine is a massive win for the community. We think this could quickly become the standard training ground for interactive and expressive robotics, significantly lowering the barrier for smaller labs trying to build natural-moving machines.