L'IA peut-elle conduire une voiture en toute sécurité dans des environnements urbains complexes ?

Waymo et Cruise se sont étendus aux zones urbaines denses (avec des revers). La barre pour « sûr » continue d'être débattue, mais la capacité technique est évidente.

Background

Current autonomous driving systems are capable of navigating complex urban environments, but their safety and reliability are still being tested and improved. Several companies, including Waymo and Tesla, have developed advanced driver-assistance systems that can handle scenarios such as intersections, pedestrian crossings, and construction zones. These systems rely on a combination of sensors—cameras, lidar, and radar—paired with high-definition maps and real-time perception algorithms to interpret dynamic surroundings. However, they may still require human intervention in rare or edge-case situations, reflecting the ongoing need for refinement. Waymo and Cruise have expanded to operate driverless in dense urban areas, though with periodic setbacks and mandated safety pauses to address incidents. Continued progress hinges on advances in sensor fusion, V2X (vehicle-to-everything) communication, and machine-learning models trained on diverse, real-world datasets. As of May 9, 2026, the Institute of Electrical and Electronics Engineers (IEEE) notes that while technical capability exists, robust, fail-safe integration across hardware, software, and cloud infrastructure remains the critical barrier to universal safe deployment in complex cityscapes.