Integration and training of a ROS autonomous driver for human-like driving style in a complex multi-component driving simulator

The advances of the last decade towards a level 4 autonomous car have been remarkable, and still the challenges to reach full automation are numerous. Amongst these, ride experience is regarded as crucial for the general acceptance of such vehicles, while assessing safety and reliability is lacking from a holistic approach. Aiming to address these two issues in parallel, we integrate a motion planner (MP) previously used on a real world car to the University of Leeds Driving Simulator (UoLDS) using the ROS messaging system. We argue that, together with the software stack developed, the UoLDS resulted in a platform suitable for development, testing and safe evaluation of MPs. Furthermore, aiming to capture a human-like driving style, the MP was trained using data coming from the driving simulator. The resulting driving style was ultimately evaluated by a number of participants at the driving simulator with encouraging results.