Adapting AUTOSAR for AI-Enabled Autonomous Driving: An Integrated Functional Safety, SOTIF, Cybersecurity, and Runtime Assurance Framework
This paper proposes the AUTOSAR-AI Safety Assurance Framework (AASAF), which integrates AUTOSAR Adaptive Platform services with ISO functional safety, SOTIF, cybersecurity, and software-update standards alongside a novel AI Assurance Tier model to enable the safe, traceable, and controlled deployment of learning-enabled components in autonomous driving systems.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are building a robot that can drive a car all by itself. You give it a super-brain made of artificial intelligence (AI), a fancy computer that learns from millions of pictures of roads, traffic, and pedestrians. This robot is amazing; it can see things humans miss and react faster than a blink. But here's the catch: this super-brain is a bit like a brilliant but unpredictable student. It might ace a test on sunny days but get confused when it rains, or it might confidently guess the wrong answer when it sees something it has never seen before. In the world of cars, getting a "wrong answer" doesn't just mean a bad grade; it could mean a crash.
To keep everyone safe, engineers have spent decades building strict rulebooks. One famous rulebook, called ISO 26262, is like a safety inspector that checks if the car's brakes and steering work perfectly every single time. Another rulebook, called SOTIF, worries about the times the car works exactly as designed but still makes a mistake because the world is too weird or complex. Then there's the AUTOSAR system, which is the universal language and operating system that lets all the different parts of a modern car talk to each other. The big question scientists are asking right now is: How do we fit this brilliant, unpredictable AI student into a car that must follow strict safety rules? If the AI gets confused, how do we make sure the car doesn't just keep driving blindly into danger?
This is exactly the puzzle Emrah Ekrem Karabag tackles in a new research article. The paper suggests that we can't just plug AI into a car and hope for the best. Instead, the author proposes a new "Safety Framework" that acts like a strict teacher sitting next to the AI student. This framework uses the existing AUTOSAR system but adds a special layer of supervision. The core idea is to separate the AI's ideas from the car's actions. The AI can suggest, "Hey, let's change lanes!" but a separate, simpler, and very predictable safety system has to say "Yes" or "No" before the car actually moves. If the AI starts acting weird or gets confused, this safety system takes over immediately and performs a "Minimum Risk Manoeuvre," which is just a fancy way of saying "slow down and stop safely."
The paper introduces a clever new way to grade how much trust we can put in different AI parts, called "AI Assurance Tiers." Think of these like levels of permission in a video game. At the lowest level (Tier 0), the AI is just a passenger giving advice, like a GPS telling you the weather. At the highest level (Tier 4), the AI is the main driver, but it is watched by a super-vigilant guard. The paper argues that just because a car part is labeled "high safety" (what engineers call ASIL D) doesn't mean the AI inside it is automatically safe. You need extra checks specifically for the AI's brain, like checking if it's looking at the right data or if it's getting too confident about things it doesn't understand.
The author also maps out exactly how to build this using the AUTOSAR tools that car companies already use. It's like giving the car a new set of instructions: "When the AI says 'Go,' the Safety Manager checks the weather, the road, and the AI's confidence score. If everything looks good, the car goes. If the AI is unsure or the road is weird, the Safety Manager hits the brakes." The paper doesn't claim this is a magic solution that solves every problem forever. Instead, it suggests a structured way to manage the risk, ensuring that even if the AI makes a mistake, the car has a backup plan that is simple, reliable, and always ready to save the day. It's about giving the AI the freedom to be smart, but keeping the safety controls in the hands of a system that never gets tired or confused.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.