Milestone has introduced A.N.N.A. โ Artificial Neural Network Agent โ a revolutionary Artificial Intelligence based on neural networks that will enable an incredible level of challenge for players, with smarter and faster opponents that will take advantage of each mistake. MotoGP 19 will be the first Milestone game where A.N.N.A. will debut, on June 6th on PlayStation 4, Xbox One and Windows PC/ STEAM. It will launch a few weeks later on Nintendo Switch.
A.N.N.A is the result of more than two years of joint development with Orobixโs engineers and data scientists, a company specialising in engineering Artificial Intelligence solutions, and itโs a new paradigm rather than traditional AI, where every action is planned by a developer in advance.
With neural network-based AI the approach is completely different; developers donโt dictate preset behaviours, they provide AI a goal and tools available to reach it. Then, through a complex rewarding system, they โteachโ AI which behaviours are useful to reach the goal and which are counter-productive; it will be the AI itself that learns which are the best actions to take to reach its goal.
This machine learning approach is called reinforcement learning and it allows one to obtain an AI that is extremely aware of the surrounding environment and consequences of its actions, differently from other approaches focused on observation and imitation of the human behaviour without a real awareness of what the AI is doing. In the first case the game strategy results from the interaction between AI and the environment, in the second one itโs imposed.
The training phase based on reinforcement learning lasts several months, during which developers can simulate more than 200,000 races in a single day, more than a professional rider can run in his entire life. In the training phase, the main activities of the Neural AI can be divided into two moments: the exploration phase, where AI understands whether the consequences of its actions are useful or not to reach its goal, and the exploitation phase, where AI take advantage of its experience and knowledge to get the best reward available.
The result is an extremely realistic and natural driving behaviour, with driving systems, manoeuvres and techniques very similar to those made by a professional rider. On the performance side, Neural AI lap times are very close to professional riderโs lap times in real life, and much faster than a traditional AI system. Behaviour in group is aggressive but fair, with AI always ready to take advantages of any mistakes. However, to match all payersโ needs, the difficulty level will be customisable. While for MotoGPโข hardcore fans, the challenge has never been harder!
โWeโre facing one of the biggest innovations in Milestoneโs historyโ said Luisa Bixio, CEO, Milestone. โThe amazing work done by our engineers, in collaboration with Orobix, is a great example of two independent companies innovating in an industry that is often controlled by big corporations. Itโs a demonstration of creativity, passion and capacity to innovateโ.
Michele Caletti, MotoGP 19 Producer, said
โwe developed the game following feedback we received from our passionate community. One of the most recurring requests were a harder challenge in single player and a more natural behaviour of the AI. With neural networks we reached both these goals, with results far above our expectations. I do think this is just the first step on an innovation path where neural networks will become crucial in our game developmentโ.
Luca Antiga, Chief Technical Officer in Orobix, said
โIt was exciting to work closely with Milestone engineers and designers and to see a new generation of AI evolving. Itโs really motivating to know that the technology we developed for many months will be on PCs and consoles all over the world. Here at Orobix weโre already working on other applications of this technology in different industries, such as medical and manufacturingโ.
