Simulation of the ADF leads to an enormous quantity of simulated kilometres. To ensure that these kilometres are worthwhile and useful, having realistic and varied virtual scenarios is extremely important. These virtual driving scenarios can be built up from the real-world traffic environment or from different driving databases (e.g. intersections, lanes, kerbs, traffic lights, pedestrians). Simulations can explore thousands of varying scenarios by applying parameter variations for the generation of novel scenarios, such as speed, trajectory or position of oncoming vehicles and the timing of traffic lights. Even the more complex scenarios need to be considered, by adding simulated traffic agents (pedestrians, joggers, motorcycles, vehicles, animals, objects, etc.) with realistic behaviours.
However, when utilising real-world data, the aspects of traceability of the data source and the influence on the result of the simulation also need to be considered (Waymo, 2018).
Main Question
Is there real driving data guiding the simulation approaches?
Sub-Questions
- Is the behaviour of the traffic agents in line with the real world behaviour?
- Are the sensor models developped based on real driving data?
- Are variations of the parameters applied in this context, and covered reality?
- Are the applied simulations based on Naturalistic Driving Study (NDS) database, accident database or records of real-world drives?
References
- Waymo (2018) On the Road to Fully Self-Driving – Waymo Safety Report. Available at: https://storage.googleapis.com/sdc-prod/v1/safety-report/Safety%20Report%202018.pdf (Accessed: 29 November 2023)