Autonomous driving refers to a vehicle’s ability to move without human intervention, thanks to a combination of sensors, software, and artificial intelligence. The topic goes far beyond laboratory prototypes: stretches of road are already open to level 3 systems in France, and the UN has adopted a global regulatory framework that changes the game for car manufacturers.
Transfer of responsibility to the manufacturer: what level 3 changes in France
The SAE (Society of Automotive Engineers) classification distinguishes five levels of autonomous driving. Level 3 introduces a break in the automotive liability regime, and this is where the legal consequences become tangible.
In France, approximately 2,000 km of roads are authorized for level 3 autonomous driving, on predefined routes around Bordeaux, Île-de-France, Strasbourg, and Isère. The activation conditions remain strict: separate lanes, absence of pedestrians and cyclists, limited speed.
The decisive point lies in the transfer of responsibility. When the autonomous system is activated in these areas, civil liability shifts from the driver to the manufacturer. This shift breaks with decades of automotive law where the driver remained primarily responsible, even in the event of mechanical failure. To delve deeper into how autonomous driving works on Actualité Premium, the topic is addressed from a technical and regulatory perspective.
The scope of use remains limited: traffic jams, certain highway sections. The driver must be able to take control at any moment if the system requests it. The actual time to regain control varies depending on the situations, and manufacturers have not yet published consolidated data on this point.
Sensors and artificial intelligence: the technical chain of an autonomous vehicle
An autonomous vehicle perceives its environment through several types of sensors that work in parallel. Each covers a weakness of the other.
- Cameras capture visual information (road markings, traffic signs, traffic lights) but lose reliability in low light or bad weather
- Radars measure the distance and speed of surrounding objects, even in bad weather, but do not finely distinguish shapes
- LiDAR sensors model the environment in three dimensions with centimeter-level precision, but remain expensive and sensitive to heavy rain or snow conditions
- Ultrasonic sensors complement the system at short range, especially for parking maneuvers
This raw data is merged and then processed by a central software, often referred to as the vehicle’s “electronic brain.” Artificial intelligence algorithms identify obstacles, anticipate the trajectories of other road users, and make real-time driving decisions.
The question of reliability over the vehicle’s lifespan remains open. The new UN regulation requires manufacturers to prove that their system detection of its environment, reacts to complex situations, and manages failures throughout the vehicle’s lifespan. This certification requirement represents a significant industrial challenge, particularly for remote software updates.
UN global regulation: an unprecedented framework for robotaxis
Until recently, each country set its own rules for the circulation of autonomous vehicles. The UN has adopted a global framework that harmonizes safety and certification requirements, with direct implications for manufacturers and robotaxi operators.
This framework requires that autonomous systems maintain an acceptable level of safety throughout the vehicle’s lifespan. Manufacturers must demonstrate their technology’s ability to handle complex scenarios: unmarked intersections, pedestrians suddenly appearing between two vehicles, degradation of a sensor on the road.
For players like Tesla, which develop their autonomous driving approach primarily from cameras (without LiDAR), this regulation creates an additional constraint. The ability of a “vision-only” approach to sustainably meet the certification criteria set by the UN remains an open question, as no certification results have been published at this stage.
Robotaxis and shuttles: the first application areas
Autonomous collective transport shuttles serve as a privileged testing ground. Their reduced speed and predefined routes simplify the task for algorithms. However, the transition to robotaxis operating at normal speed in dense urban environments remains a technological threshold that few operators have sustainably crossed.
The automotive industry faces a trade-off: invest heavily in developing level 4 and 5 systems (full autonomy), or consolidate level 3 by gradually expanding geofenced areas. The majority of European manufacturers favor this second approach, deemed more realistic in the medium term.
Road safety and autonomous driving: available data
The central argument in favor of autonomous driving is based on a finding: the vast majority of road accidents result from human errors. Eliminating the human factor should, in theory, drastically reduce road fatalities.
Initial feedback from real deployments nuances this projection. Autonomous systems eliminate certain types of accidents (distraction, drowsiness, alcohol) but introduce others, related to scenarios that the algorithm has not learned to handle. Cases of unjustified sudden braking or hesitation in the face of an unusual obstacle have been documented by several operators.
The question of coexistence between autonomous vehicles and human drivers on the same roadways adds a layer of complexity. An autonomous vehicle strictly adheres to traffic laws, which can create unexpected situations for human drivers accustomed to more “fluid” behaviors (lane changes, acceleration during merging).
The development of autonomous driving in France and worldwide progresses through regulatory and technological stages rather than through a sudden break. The coming years will determine whether the expansion of level 3 areas and the arrival of level 4 on certain routes live up to their promises in terms of safety, without the current legal framework, still under construction, hindering adoption by drivers.