NVIDIA and Google DeepMind have teamed up to launch Newton, an open-source engine that’s designed to close the gap in advanced robotic systems. Using NVIDIA’s Warp platform, Newton is able to deliver a scalable performance that is designed to provide simulations that support real-life physics.
Incorporating Physics and Real-Life Actions
Even though it’s been possible to recreate a number of elements from our world in a digital format for quite some time, replicating physics has always been a challenge. With that said, notable progress has been made.
For example, Kip Thorne, a theoretical physicist used actual equations based on Einstein’s general relativity to make the movie, Interstellar, more realistic. He also used equations to work out how a black hole would bend light, and change over time. In games, like Fishing Planet, physics engines are used to determine the trajectory of the bait or lure, based on motion, wind, power and angle. Even in the Plinko casino game real money, game engine physics are used to recreate the natural path a ball would take when dropped for a height. The amount the ball bounces off each peg also comes down to physics, with users able to change the size of the board to cater the experience. Thanks to advancements, this can all be done while abiding by the laws of physics when playing on a real board.
Of course, as time goes on, more and more work is being done to understand physics, so that our real world can be better replicated in a digital format, and with the latest announcement regarding Newton, it seems that outstanding progress is being made which is set to kick things up a notch for the tech sector.
Popular Robotics and Frameworks
Newton uses CUDA-based GPU libraries to create physics simulations. It integrates with popular frameworks including Isaac Labs to try and help developers work on humanoid systems. One major feature here would be the MuJoCo-Warp. This is a high-speed extension that is able to speed things up by over 700%. It also provides 1000% faster acceleration for things like hand manipulation, which is said to reduce the amount of time it takes to utilise AI training.
The main aim of this development is to try and reduce the sim-to-real gap. This is the disparity between real-world performance and simulated environments. Even though the support is for differentiable physics, Newton is able to translate simulation learning to real-world behaviours as well. The fact that it is so easy to work with also makes it a solid foudnation for the future.
This means that developers can incorporate their own ideas and simulate more actions, whether it is by using deformable materials like sand or cloth, or interacting with soft elements. This kind of flexibility makes it ideal for a huge range of applications that go far beyond the world of humanoid robotics, and it also helps to pave the way for many new exciting developments within the sector.
With tech news like this set to change the world of digital physics as we know it, it’s only a matter of time before we could be seeing even bigger developments that bridge the gap between the real and online world.
