A couple of days ago, the classic arcade game PAC-MAN was a whopping 40 years old Since its first arrival at Japanese arcades and arcades, a milestone followed by an impressive expansion that made it one of the most iconic and representative titles in the entire video game industry. And it is that in all these years we have seen a multitude of versions and deliveries, to which today this curious adaptation of NVIDIA, generated through a simulated game engine created through Artificial Intelligence.
Trained with over 50,000 games, this new AI created by the NVIDIA Research team, GameGAN, has been shown to be able to generate a fully functional version of PAC-MAN without an underlying game engine, keeping track of the virtual world and remembering what has already been generated to maintain visual frame-by-frame coherence. That means, even without understanding the fundamental rules of a gameAI can recreate the game with compelling results.
It’s like this of the first neural network model that mimics a computer game engine by leveraging adverse generative networks, or GAN. Thus, made up of two opposing neural networks, as a generator and a discriminator, these artificial intelligence models learn to create new content that is convincing enough to go through the original PAC-MAN.
In this way, and through previous training with game scripts and multiple levels, GameGAN could become even capable of generating new game designs never seen before, and fully functional for the gameplay and perception of the market and current players.
Thus, in addition to opening a new world of possibilities for developers such as automatic generation of designs and new game levels, we are faced with a new AI capable of developing more easily simulation systems to train autonomous machines.
From games to real life
Autonomous robots are usually trained in a simulator, where AI can learn the rules of an environment before interact with objects in the real world. However, creating these simulators goes through a slow and tedious process in which you have to code the rules on how objects interact with each other and how light works within the environment.
Simulators are used to develop autonomous machines of all kinds, such as warehouse robots that learn to grab and move objects, or delivery robots that must navigate the sidewalks to transport food or medicine.
What years ago seemed an idyllic future only possible in some science fiction short films or video games, has turned out to be part of our day to day, so much so that in fact, they have become some of the main guidelines that have marked the next steps of our technological future.
And it is that without being chased by ghosts, today there are numerous companies like Amazon, AliExpress or Correos, which already use autonomous robots for the management of their warehouses, following rules quite similar to those of this classic arcade.