Physics simulation trains robots 430,000 times faster
Briefly

According to Jim Fan, co-author of the Genesis paper, the speed of simulation is revolutionary: "One hour of compute time gives a robot 10 years of training experience." This acceleration allows for extensive and rapid learning in a virtual world.
Genesis was developed to empower researchers as they experiment with robots in simulated environments, ultimately reducing the financial burden of physical testing by making learning faster and more efficient.
Zhou Xian, leading the team at Carnegie Mellon, highlighted Genesis's capability: "The platform processes physics calculations up to 80 times faster than existing simulators, allowing for more complex and realistic robot training scenarios."
Fan elaborated on the potential scale of training, stating, "If an AI can control 1,000 robots to perform 1 million skills in 1 billion different simulations, it may 'just work' in our real world."
Read at Techzine Global
[
|
]