The Japanese Fanuc and Nvidia announced an integrated system where robots trained in virtual environments begin to behave in practically the same way when they are transferred to real factories and perhaps this seems like just a technical detail until you understand the impact it has on global automation. Today, training and validating industrial robots is usually an extremely expensive and slow process.

Companies need to physically test machines, adjust parameters in the real environment, deal with unexpected mechanical errors, and repeat validation cycles countless times. The new system attempts to eliminate much of that. The integration combines Nvidia ISXM with Fanuc's Robot Guide industrial software to create extremely accurate digital twins of factories.

Within these simulated environments, robots use exactly the same control algorithms as in real machines, that is, the virtual robot learns in a digital world, but executes in the physical world in almost the same way. And this changes things quite a bit because now engineers can validate extremely complex tasks before even turning on the real robots.

The system manages to simulate cable manipulation, fitting of parts, industrial assembly and even unpredictable situations with randomly dispersed objects, all using simulated physics in real time and the most interesting detail seems to be in the increasing focus on tasks that were previously considered too human for efficient automation.

Fanuc itself recently demonstrated robots learning to fold t-shirts using learning by imitation, a human operator initially performs the movement while the robots observe and learn the behavior and perhaps this is precisely where this technology begins to seem important beyond factories, because the moment robots manage to learn in virtual environments extremely faithful to reality, training stops depending solely on the physical world. This drastically accelerates the evolution of machines, and even humanoid machines that are also entering the agricultural sector.