Porsche is looking at using artificial intelligence (AI) to enhance its driver-assistance systems and try to tackle one of the biggest challenges in autonomous driving: dealing with rare, unpredictable traffic situations. If it really works, WOW.
The Problem with Traditional ADAS
Advanced driver-assistance systems have certainly come a long way, but they still do not handle the unusual scenarios as well as a human would. These systems are based on millions of miles of driving data fed into machine learning models, but rare edge cases—like an animal darting onto the road, an unusual road sign placement, or even a flat-bed truck carrying a car—can confuse them.
For instance, one of the most common failures in ADAS is when the system interprets a flat-bed truck carrying a vehicle as a car going the wrong way. Another common failure is the interpretation of snowy verges as lane markings and thus causing wrong driving inputs. Since these cases are very rare, training the ADAS for such scenarios is a major challenge.
How Porsche's AI Solution Works
Porsche Engineering has developed a new AI-based approach to tackle the problem. The system works based on a type of artificial neural network called Variational Autoencoder (VAE), designed to process large amounts of sensor and camera data. The AI then learns from this data, identifies patterns within it, and distinguishes routine driving situations from rare anomalies.
The VAE works by compressing complex visual and sensor inputs into simpler representations. This helps the AI model focus on rare and unusual incidents instead of being overwhelmed by everyday driving data. In simple terms, the AI filters out the noise and focuses on the real challenges.
Why This Matters for the Future
The main advantage of Porsche's AI technology is that it allows ADAS to continuously learn and improve, thus helping handle more unexpected road conditions effectively. Instead of relying on predefined rules, the AI adapts to new data, thus enhancing its ability to respond more effectively in the event of rare traffic incidents.
Currently, Porsche's AI system is working with recorded data stored in the cloud. However, the long-term goal is to integrate this technology into real-time driving assistance systems, making future Porsche models safer and smarter.
The Road Ahead
“The capabilities of the method are currently being tested worldwide. This allows us to identify very different types of corner cases and also to draw conclusions about the extent to which the results can be transferred from one country to another,” says Arsen Saigoian, Specialist Project Manager in the AI and Big Data division at Porsche Engineering.
