High-Level Decision-Making Non-player Vehicles.

Availability of realistic driver models, also able to represent various driving styles, is key to add traffic in serious games on automotive driving. We propose a new architecture for behavioural planning of vehicles, that decide their motion taking high-level decisions, such as "keep lane", "overtake" and "go to rightmost lane". This is similar to a driver's high-level reasoning and takes into account the availability of ever more sophisticated Advanced Driving Assistance Systems (ADAS) in current vehicles. Compared to a low-level decision making system, our model performs better both in terms of safety and average speed. As a significant advantage, the hierarchical approach allows to reduce the number of training steps, which is critical for ML models, by more than one order of magnitude. The developed agent seems to show a more realistic behaviour. We also showed feasibility of training models able to differentiate their performance in a way similar to the driving styles. We believe that such agents could be profitably employed in state of the art SGs for driving, improving the realism of single NPVs and overall traffic.