Given its complexity, the testing of level 3 and 4 highly automated driving (HAD) has become the bottleneck. The old paradigm of mileage-driven testing will need to be
replaced by scenario-based testing to test autonomous vehicles in a measurable, consistent and deterministic way. The new paradigm, scenario-driven development
is most efficient when adopted early in the process of design & development of HAD.
In this presentation, we define scenarios and motivate why they are important. We illustrate at what points in the value chain of autonomous vehicle design & deployment scenarios create value. We present our pipeline for scenarios which consists of the following 4 steps: Firstly, we do scenario identification to determine what is relevant and what should be simulated. Secondly, we automatically extract observations from multiple sources and highlight the advantages of observation-based extraction over scenario creation. Thirdly, the important step of fuzzing and pruning by combining relevant scenarios in a meaningful way. Different use cases and ODDs require different fidelities and different simulation variants. We end our presentation by giving a live demonstration of our scenario pipeline.
- Scenario definition and motivation – We define scenarios and motivate why they are important. We illustrate at what points in the value chain of autonomous vehicle design & deployment scenarios create value.
- Furthermore, we highlight the key role scenarios play in reaching higher levels of autonomous driving.
- Scenario identification – how to determine what is relevant and what to simulate.
- Scenario extraction – automatic observation-based extraction from multiple sources and the advantages of observation-based extraction over constructed scenarios.
- Fuzzing / Pruning of scenarios – combining relevant scenarios in a meaningful way.
- Simulation (variations) / different fidelities for different use cases.