A team of researchers from universities and industry, led by Zu Xian from Carnegie Mellon University, has launched “Genesis,” an open-source platform that is creating a paradigm shift in robot training using virtual simulation. This platform allows robots to simulate training at a blazing speed of up to 430,000 times faster than real-time, saving years of learning in just a few hours.
Faster and more efficient training
Genesis uses advanced technology that enables robots to practice tasks in virtual simulation environments with unprecedented accuracy and speed. The neural networks for robots can practice picking up objects, walking, or using tools at a rate equivalent to decades of real-world training, but within just a few hours of computing time.
Jim Fan, one of the co-developers of the project, said: “One hour of compute time is equivalent to 10 years of robot training experience. This is like how Neo learned martial arts at lightning speed in the movie The Matrix.”
Innovation in simulation design
The Genesis platform features the ability to perform physics calculations 80 times faster than existing simulators like NVIDIA’s Isaac Gym. The platform uses ordinary graphics cards to run up to 100,000 instances of a single simulation simultaneously. This innovation is crucial for training the neural networks that will control future robots.
This new approach represents a major step forward in improving robots’ ability to learn through virtual environments that accurately mimic reality, reducing the need for costly and complex real-world testing.
Creating dynamic virtual worlds
One of the key advantages of the Genesis platform is its ability to create dynamic 3D virtual worlds known as “4D worlds” because they combine three dimensions with time. Using language and vision models, the platform can generate complex virtual environments through text commands alone.
This technology enables researchers to create robot testing environments with high accuracy and speed without the need for complex programming or manual design. Fan wrote: “Creating simulations typically requires a massive effort to design 3D assets and set up scenes, but with Genesis, the process can be fully automated.”
An open-source platform for everyone
What sets Genesis apart from other simulators is that it is an open-source platform, making it available for free to any researcher or developer who wants to use it. The platform relies entirely on Python for its interface and physics engine, making it easy to use widely without the need for complex setups or software.
Through this platform, anyone can run robot simulations quickly and efficiently using ordinary computers. Fan emphasized in his announcement that robotics should be a collaborative project among all humans, rather than an exclusive domain for major corporations.
Multiple future applications
In addition to robot training, Genesis offers broad capabilities that include generating character movements, interactive 3D scenes, and even facial animations. These features can be used in developing digital games and future films, opening up new horizons for digital creativity.
Although the dynamic world generation feature is not currently available within the open code on GitHub, the team is working to release it soon.
Scientific community support
The Genesis platform provides an opportunity for the scientific community to participate in its development and improvement through its open GitHub repository. Researchers and developers can contribute to enhancing the system and adding new features, making it an advanced and accessible tool for everyone.
Toward the future of artificial intelligence
Genesis represents an important step toward developing artificial intelligence technologies and training robots quickly and efficiently. Thanks to being an open-source platform, everyone can benefit from it to accelerate research and development in this promising field, contributing to the creation of a more advanced and inclusive digital world.
