Artificial Neural Expert System based YOLO$_{Gen_1}$ detector for Autonomous Driving Problem

Driving a car requires a constant, split-second conversation between the driver and the road. The eyes scan for hazards, the brain processes the meaning of a flashing light or a sudden stop, and the hands execute the necessary turn or brake. For decades, engineers have tried to give machines this same ability, building systems that can see and decide without human help. The challenge is immense because the world is unpredictable; a pedestrian might step off a curb, or a traffic light might malfunction. To solve this, researchers rely on two main tools: sensors that act as the machine's eyes, and software that acts as its brain. The eyes usually use cameras or radar to spot objects, while the brain uses complex computer programs to figure out what those objects mean and what to do next. The goal is to create a vehicle that can navigate safely and efficiently, handling the messy reality of real-world traffic.

A researcher at Tunis El Manar University has proposed a new way to build this system, combining a sharper set of eyes with a more experienced brain. They introduced a new detector called YOLOGen1, which serves as the visual sensor for their autonomous driving system. Unlike previous versions that often struggled with small or distant objects, this new detector is designed to see the road with exceptional clarity. It does not rely on radar, which can be affected by weather, but instead uses a sophisticated combination of camera-based models working together. The researcher found that this new detector could identify objects with a confidence level exceeding 98 percent on standard driving datasets. In tests using a specific dataset of driving scenes, it outperformed several other leading models, finding more cars, pedestrians, and signs with greater accuracy.

However, seeing the road is only half the battle; the vehicle must also decide how to react. The researcher addressed this by creating a new type of decision-making system they call an Artificial Neural Expert System. Think of this as a digital driver who has studied the rulebook of driving but also learns from their own mistakes. Traditional computer programs follow rigid rules: if the light is red, stop. But what if the light is broken and stuck on red? A rigid program would stop forever. The new system, however, can recognize that the light is blocked and switch its strategy, perhaps treating the situation like a stop sign instead. It does this by constantly updating its internal rules based on what happens next, allowing it to handle situations where the standard rules fail. This system is designed to be safe, ensuring that even when it learns, it never chooses a dangerous path.

To make this system work in real time, the researcher organized the software so that different parts of the computer work together simultaneously, rather than waiting for one task to finish before starting the next. They tested this entire setup on a standard laptop processor, without needing expensive graphics cards. The results were striking: the system could process video and make driving recommendations faster than real time, even on simple hardware. In simulations, the system successfully navigated complex scenarios, such as a traffic light that remained red due to a malfunction, by dynamically changing its decision logic to keep the vehicle moving safely. The researcher validated their approach using real-world video footage, showing that the system could provide useful, immediate advice to a driver or an autonomous vehicle.

The study suggests that by combining a highly accurate visual detector with a flexible, learning-based decision system, autonomous vehicles can become safer and more reliable. The researcher demonstrated that their new detector, YOLOGen1, is currently the most effective version of its kind for spotting objects in driving scenes. They also showed that their decision system can adapt to broken rules and unexpected obstacles, a critical step toward vehicles that can handle the chaos of real traffic. While the system has been tested extensively in simulations and with recorded videos, the researcher notes that further testing on actual roads is necessary before it can be used in production cars. Their work offers a promising path forward, proving that a machine can not only see the road clearly but also think through the problems it finds there.