Everything your parents didn’t tell you about AV and ADAS needs, from Level 2 to Level 5.
19th Sep 2023 dRISK
By Kiran Jesudasan
Autonomous Vehicles are inherently complex, and their data requirements are multifaceted. For the purpose of this blog article, our focus will be on the evolving regulatory landscape of testing. We will delve into the history of regulatory testing, highlight key changes projected for the industry, and demonstrate how dRISK aligns directly with this evolution. Specifically, we will examine the testing of automated features and how it has evolved.
The European National Crash Assessment Program, commonly known as EuroNCAP, was established in 1996. It was the result of collaborative efforts between various departments of transport and safety groups across Europe, aiming to create standardized testing procedures. The idea behind a centralized testing system was to eliminate the need to pass individual country’s testing procedures, simplifying the process and facilitating market entry. Thus, EuroNCAP was established. In 2001, they awarded their first 5-star rating to the Renault Laguna.
2009. EuroNCAP began testing Electronic Stability Control (ESC) by subjecting vehicles to rigorous turns in wet weather conditions, comparing performance with and without the system activated to assess their ability to stay on track. However, after ESC became mandatory in 2014 and virtually every car was equipped with it, EuroNCAP discontinued testing this feature as it no longer provided a competitive advantage.
2014. EuroNCAP initiated their testing protocols for Automatic Emergency Braking (AEB), marking the beginning of automated safety feature testing. The initial tests for AEB were relatively straightforward, consisting of two types: one where an ego vehicle (the car under test) approached a stationary car, and another where it approached a moving car at a consistently slower speed.
2015. EuroNCAP introduced tests for Automatic Emergency Braking (AEB) performance involving pedestrians. These tests consisted of two types: the first involved an adult manikin walking perpendicular to the path of the ego vehicle, while the second test replicated the scenario with a child manikin. Notably, the latter test incorporated additional complexity by introducing two additional vehicles as occlusions. This marked the first instance of enhancing testing regimes to assess the effectiveness of specific autonomous features. It’s an important trend to observe as it showcases the evolving nature of testing methodologies.
2018. EuroNCAP started to test AEB with cyclists, again with two variations where the cyclist is moving perpendicular to the vehicle under test and another where both travel in the same direction. Start to see a trend emerging?
In the same year, EuroNCAP introduced additional tests to assess level 2 functionality. These included various scenarios for AEB, such as evaluating the ego vehicle’s response when another vehicle cuts into its path, when a vehicle in front of the ego suddenly veers out of the way to reveal a stationary car ahead, and how the ego vehicle reacts to a decelerating vehicle in front of it. Furthermore, the testing regime incorporated navigating an S curve to evaluate the performance of Automatic Lane Keeping Systems (ALKS). Lastly, the assessment also measured how well the ego vehicle avoids obstacles placed on the road.
2020. EuroNCAP further broadened its testing scope. They incorporated scenarios such as unprotected turns involving both pedestrians and vehicles, AEB with pedestrians while reversing, and assessments that focused on pedestrian safety during nighttime conditions. It becomes evident that there is a clear trend toward expanding the testing of active safety features in conditions that mimic real-world conditions.
2022. The European Commission starts applying the new General Safety Regulation (GSR 2), which makes many active safety and advanced driver assistance features mandatory for all cars sold in the EU. Similar to the evolution of EuroNCAP testing, the GSR 2 requirements start with relatively mature features, such as AEB, but also paves the way towards having fully autonomous and driverless cars in the EU.
When regulators launch a test program, it typically begins with relatively simple tests. As time passes and more data becomes available, along with pressures from the industry, consumers, and other regulatory agencies, testing mandates become more complex. In the case of ADAS and ADS testing, it started with a few scenarios involving car-to-car interactions and then expanded to include pedestrians and cyclists, and eventually objects. Initially, testing took place in clear environmental conditions, but now it is also progressing towards nighttime testing. The aim is to continuously develop tests that accurately capture the complexity of the real world and the myriad of peculiar and fascinating phenomena we encounter daily on or near our roads.
What the market needs is:
- An understanding of all of the GSR2 requirements and how they are going to expand
- Scenario data that enables manufacturers to test GSR compliance of active safety features early in the development cycles
- Tools for managing test data that adapt to more and more complex regulatory requirements
- Knowledge of how testing frameworks differ from market to market, down to the level of concrete testing procedures and pass-fail requirements
- A clear vision of how these testing requirements will evolve for level 4 and 5 systems
We developed dRISK Edge as a data management platform that enables users to visualize and explore the entirety of testing and training scenarios for ADAS and AVs. These scenarios include test cases for active safety features mandated by GSR 2, which are represented by a cluster of green nodes in the above screen recording.
Orange scenarios are not mandated by GSR2 yet but are specified based on the evolution of the EuroNCAP tests. The embedding space plotted here is a convenient method for exploring clusters of similar scenarios. The videos of simulation renderings attached to the scenario nodes allow further interrogation and show how the advanced scenarios differ in terms of complexity.
Red nodes are specified based on where EuroNCAP is expected to move towards in 2025 and 2030, i.e. capturing more real-world complexity. For example, these scenarios include a pedestrian crossing the path of the ego in a roundabout, not just a pedestrian and a roundabout in separate scenarios.
dRISK customers have access to scenario data that can be used in different simulation platforms (e.g. IPG Carmaker, NVIDIA DRIVE Sim, rFpro) for virtual testing against the current EuroNCAP and GSR 2 requirements. But the dRISK Knowledge Graph also contains many more scenarios that are not (yet) relevant to regulated or voluntary safety assessment programmes, including many edge cases. Ultimately, dRISK aims to help manufacturers build crashless cars that can handle anything the world throws at them, and EuroNCAP and GSR 2 are only the first steps toward this goal.