Before founding Vector Anomaly, we were employed at Microsoft's Cambridge Research laboratory where, in conjuction with Microsoft Games Studios, we developed the Drivatar* artificial intelligence (AI) system for the Xbox racing game Forza Motorsport.
Players of Forza will most readily associate the term 'Drivatar' with the trainable Driving avatar entity introduced by the "Train Drivatar" mode in the game. In fact, the Drivatar AI system extends considerably beyond that novel gameplay concept. The cutting-edge machine learning research technology that underpins the "Train Drivatar" feature was also the foundation of all the built-in AI opponents in Forza.
So, while the core concept of a Drivatar is that they are able to continuously learn a particular driving
style from a human teacher, they can also be taught ahead of time by appropriately-constructed development tools.
Whatever method is used to create a Drivatar, the result is a probablistic model of the desired driving style. The probabilistic nature of the Drivatar implies that it will demonstrate driving behaviour which is realistically variable and unpredictable, yet which is nevertheless visibly consistent with the
appropriate style of driving — either that of the human teacher, or that pre-specified for the built-in AI.
While inevitably we must be somewhat guarded in what information we can reveal, in these pages we offer a mixture of general thoughts on racing game AI, further details of the Drivatar AI system and some nuggets of information regarding its implementation in Forza Motorsport. Simply follow the links below to find out more, and thanks for reading!
|
