2024-08-19 06:30:56
In the early 2000s, it was thought that our roads would be loaded with fully autonomous vehicles by 2020. In 2024, the reality is very different! While there are vehicles offering a certain level of autonomy, no vehicle has yet been approved to drive without supervision. And the growth of these vehicles depends largely on the testing methods that will be put in place for this approval.
There are five levels of autonomy for vehicles. We only start talking about automated driving from level 3, level 1 being assisted driving (cruise control, anti-lock braking system, etc.) and level 2 being partial autonomy (parking assistance, trajectory correction, etc.).
Autonomous vehicles are already authorized in France!
France is one of the most advanced countries in Europe on the issue of autonomous vehicles, a legislative framework having been adopted on September 1, 2022.
Does this legislative framework imply that autonomous vehicles are authorized in France? This is indeed the case! At least, in theory, and under very specific conditions. Because to be authorized to circulate:
- the vehicle must be on a road with physical separation of opposing lanes, where pedestrians and cyclists are prohibited;
- its speed is limited to 60 km/h;
- the driver must be in a position to take over at all times.
Furthermore, very few approved and marketed models are currently available to drivers who might be tempted by this type of vehicle. Indeed, in the world, only two level 3 models have received approval: the Mercedes class S in Germany and the Honda Legend in Japan.
On the other hand, although experiments are underway, no level 4 model has yet been approved and as for level 5, it is still at the idea stage!
Safety: the main obstacle to the growth of autonomous vehicles
Operating autonomous vehicles carrying passengers in high-traffic areas poses obvious safety concerns, whether the vehicles are on land or even in the air.
Technological feasibility is therefore not the main obstacle, especially since recent progress in artificial intelligence makes it possible to consider driving vehicles autonomously in increasingly complex environments.
Currently, we know how to operate autonomous vehicles in a closed environment, with known and controlled conditions. But operating these vehicles in an “open” environment while guaranteeing an acceptable level of safety without falling into excess safety (untimely braking) is far from obvious.
Furthermore, demonstrating the safety of these vehicles in real conditions by taking into account an infinite number of situations is impossible in practice. Especially since these vehicles would have to be tested over billions of kilometers!
Digital simulation is a valuable tool, in addition to road tests
Today, the behavior of autonomous vehicles is tested by restricting their Operational Design Domain (ODD), in other words their operating ranges: type of road, traffic, defined brightness, etc. But how can we ensure that the vehicle’s behavior remains safe outside of these specific conditions?
Digital simulation then appears as a complementary solution to guarantee the exhaustiveness of test conditions, without causing the costs and time required to carry out the tests to explode.
In France, various projects of this type are underway, involving key players such as the National Metrology and Testing Laboratory (LNE), the CEA and the IRT SystemX. We can mention the 3SA project, currently being finalized and which brings together these three players as well as several industrialists including Renault and Stellantis.
The objective of 3SA is to develop simulation methods and tools involving sensor modeling, the provision of a library of scenarios, the development of a method for analyzing simulation results and the creation of a methodological reference framework.
According to Rémi Régnier, Research Engineer in the Artificial Intelligence Evaluation and Cybersecurity Department at LNE, « In this context, the LNE has developed references, defining what the appropriate responses of an autonomous vehicle should be, as well as an evaluation framework to establish whether the test is valid or not. All this, while ensuring total coverage of the ODD, and working both on the robustness of the systems, i.e. their stability and resistance to environmental disturbances, and their resilience.
The LNE is also involved in other projects, including:
- PRISSMA[1]which aims to create methodologies for approving AI components for autonomous vehicles;
- TEF[2] (Testing and Experimentation Facilities), the aim of which is in particular to make testing methods and means accessible to the European industrial sector to qualify the performance, reliability and robustness of AI systems.
The democratization of autonomous vehicles is therefore not happening right away, but one thing is certain: it is mainly based on the implementation of testing methods adapted to their approval, which are still in the field of research.
[1] Supported by the Ministry of Transport, supervised by UTAC and Gustave Eiffel University (UGE)
[2] Funded by the EU and Member States and through the National Strategy for AI, for France
1724067886
#Autonomous #vehicles #developing #testing #methods #condition #approval
