In recent years, autonomous driving systems in cars have been developing rapidly, particularly in models reaching L2+ and L3 levels, where dependence on passenger intervention is decreasing. So, how reliable do you think these systems are in real-world conditions? With increased automation, how practical is it for systems to be taken over instantly by humans? In terms of safety, what are your thoughts on manufacturers' sense of responsibility and the adequacy of regulations?
How safe is L2+/L3 level autonomy in cars?
👁️ 8 views💬 2 replies❤️ 0 likes
2 Replies
I still can't find the 'D' button on an automatic gearbox when they're talking about L2+/L3 stuff 😂. How can passengers' reflexes work before the system disables me? I'm running from a chicken—how can they guarantee the sensors will run from the chicken too? 🦆💨
Level 2+ and Level 3 autonomous systems fall into what could be described as "partial autonomy" or "autonomy under human supervision." In L2+, the system handles both steering and acceleration/braking, but still requires constant human attention. In L3, the system can take full responsibility in certain situations, but provides a "reasonable time" for human intervention. In real-world testing, the biggest safety risk arises precisely during these transition moments: the system issues a sudden alert, and the human may not react as quickly as the algorithm expects.
Even systems like NVIDIA's DRIVE platform or Tesla's FSD, despite their high computing power, have limitations in environmental awareness—especially in poor weather conditions, sensor fusion, or rare scenarios. For example, there's always a risk of failing to respond to a vehicle stopped on the road due to glare or misclassifying a cyclist on the side of the road. While manufacturers try to clarify system limitations to shift responsibility, legal and ethical issues persist. The "practicality" of human intervention depends on individual reflexes; taking over the wheel in an emergency, especially during long journeys with physiological fatigue, can be challenging.
In reality, the most critical aspect of L3 systems is how well manufacturers optimize what's called the "courtesy time" for transitions. Standards like the EU's UNECE R157 regulation already set minimum safety criteria for these transitions, but real-world testing still shows variations. From a safety perspective, I believe the most reliable approach is to stay with L2+ and maintain human oversight throughout the process. Transitioning to L3 will only make sense once the algorithm's reliability and the human-machine interface have matured.