Newton is an open-source physics engine developed by NVIDIA, Google DeepMind, and Disney Research to accelerate robotics development. This powerful tool enables 70x faster robot learning compared to existing solutions by combining GPU-accelerated simulation with differentiable physics, allowing robots to master complex tasks like object manipulation and dynamic movement with unprecedented precision .
What Makes Newton Different from Other Physics Engines?
Newton stands out through its:
-
GPU acceleration via NVIDIA Warp framework
-
Compatibility with MuJoCo and Isaac Lab
-
Differentiable physics for gradient-based optimization
-
OpenUSD integration for standardized workflows
-
Disney’s proprietary physics enhancements
Chart: Newton vs. Traditional Physics Engines
| Feature | Newton | Traditional Engines |
|---|---|---|
| Speed | 70x faster | Baseline |
| GPU Acceleration | Yes | Limited |
| Differentiable | Yes | No |
| Open Source | Yes | Mixed |
How Will Disney Use Newton for Next-Gen Robotics?
Disney plans to deploy Newton-powered robots like the BDX droid across theme parks, enabling:
-
Expressive character movements
-
Natural object interactions
-
Safe human-robot coexistence
-
Dynamic performance adjustments
What Technical Innovations Power Newton?
Key technical breakthroughs include:
-
Warp-accelerated contact dynamics
-
Multi-physics coupling (solids, fluids, fabrics)
-
Gradient-based policy optimization
-
Unified OpenUSD representation
Chart: Newton Performance Benchmarks
| Task | Speed Improvement |
|---|---|
| Humanoid Control | 70x |
| Manipulation | 100x |
| Cloth Simulation | 45x |
Buying Tips
For developers exploring Newton:
-
Verify GPU compatibility (NVIDIA RTX 3000+ recommended)
-
Prepare for Q4 2025 open-source release
-
Consider Isaac Lab for full robotics stack
-
Evaluate MuJoCo-Warp for transitional workflows
Fly-Wing Technology offers compatible hardware with global logistics support for robotics developers adopting Newton-based solutions.
Electronic Components Expert Views
“Newton represents a paradigm shift in robotics simulation by finally bridging the reality gap between virtual training and real-world deployment. Its differentiable physics will enable breakthroughs in reinforcement learning that were previously computationally impractical.” – Senior Robotics Architect
FAQ
Q: When will Newton be publicly available?
A: NVIDIA plans an early open-source release in late 2025.
Q: What hardware is required?
A: CUDA-enabled NVIDIA GPUs with at least 8GB VRAM .
Q: Can Newton simulate soft bodies?
A: Yes, including food, fabrics and granular materials.
Q: Is Newton compatible with ROS?
A: Yes, through Isaac Lab integration .
In today’s rapidly evolving landscape of artificial intelligence and robotics, NVIDIA, Google DeepMind, and Disney Research have jointly launched Newton, an open-source physics engine set to completely transform robotic learning and development. Newton’s introduction signals a significant leap forward in robotics technology, opening up new possibilities for robotic applications across various fields, from industrial settings to entertainment.
Core Technology and Performance Benefits
At the heart of Newton lies the NVIDIA Warp framework, a CUDA-based acceleration library that allows for the writing of high-performance, GPU-accelerated code directly in Python. By leveraging the parallel processing power of NVIDIA GPUs, Warp enables roboticists to train models at unprecedented speeds. This capability allows Newton to perform real-time simulations, enabling robots to refine their movements with greater precision, adapt to complex environments more effectively, and excel in both industrial precision tasks and interactive entertainment roles.
In terms of simulation speed, Newton achieves a major breakthrough through its integration with MuJoCo-Warp. Developed by Google DeepMind, MuJoCo-Warp combines the physics engine of MuJoCo with the acceleration capabilities of Warp, resulting in simulation speed improvements of over 70x for robotic manipulation tasks, with certain specific tasks experiencing up to a 100x speedup. This advancement means that complex physical interactions, such as a robotic hand manipulating a tool, can be simulated with exceptional efficiency, significantly reducing the development cycle and computational costs of robotic systems.
Innovative Differentiable Physics
Newton introduces the innovative concept of differentiable physics, breaking away from the limitations of traditional one-way physical simulations. Unlike conventional physics engines that merely generate outputs based on inputs, Newton utilizes backpropagation to optimize system parameters. This allows robots to continuously improve their interaction with the environment, similar to how deep learning models adjust their weights. This capability is crucial for the development of autonomous systems capable of self-improvement, enabling robots to enhance their performance in fields such as manufacturing, logistics, and healthcare through ongoing learning processes.
Transforming Entertainment Robotics
In the realm of entertainment robotics, Disney Research is integrating Newton into its robotic character platform, breathing new life into next-generation entertainment robots. The BDX droids, inspired by Star Wars and showcased during NVIDIA’s GTC keynote, are the first成果 of this platform. These robots, empowered by Newton’s physics simulation capabilities, exhibit lifelike movements and interactive capabilities, delivering an unprecedented entertainment experience to audiences.
Kyle Laughlin, Senior Vice President of Walt Disney Imagineering Research and Development, expresses confidence in Newton’s potential: “The BDX droids are just the beginning. We are committed to bringing more characters to life in ways the world hasn’t seen before, and our collaboration with Disney Research, NVIDIA, and Google DeepMind is key to realizing this vision. This partnership will enable us to create a new generation of robotic characters that are more expressive and engaging than ever before, connecting with our guests in ways that only Disney can.”
Unified Robotic Workflow
Newton is also built on the OpenUSD (Universal Scene Description) framework, a powerful tool that enables seamless data aggregation and simulation. OpenUSD provides a common language for robotic simulations, allowing data to be easily shared and replicated across different platforms. In collaboration with partners such as Google DeepMind, Intrinsic, and NVIDIA, Disney Research is defining an OpenUSD asset structure tailored for robotics. This standardized approach will unify the workflows of roboticists worldwide, accelerating progress in the field.
Future Outlook
As Newton continues to evolve, it is poised to become a cornerstone in the advancement of humanoid robotics. The first version, scheduled for release later this year, will usher in a new era where AI-driven, high-speed robotic simulation becomes the norm. The collaboration among NVIDIA, Google DeepMind, and Disney Research not only introduces new tools and methods to robotics but also paves the way for the widespread application of robots in the real world. Newton’s open-source nature will inspire creativity among researchers and developers globally, driving the field of robotics toward new heights.
