se, thereby enabling appropriate countermeasures to be taken. In a sense, we're embarking on a journey to a "conscientious" technology.In their development of assistance systems, researchers are also making use of the possibilities offered by the DaimlerChrysler driving simulator. Here, assistance functions can be "tested" long before a test vehicle actually hits the road. "Realistic" and even "dangerous" test drives can be conducted with any vehicle type desired. An important aspect here is determining the extent to which assistance systems are understood and accepted in borderline situations.The basic building blocks of future assistance systems that will take their surroundings into account include the ability to recognize the lane, not only through the lane markings but also through the recognition of the free area in front of the vehicle. Another important factor is the ability to recognize pedestrians, vehicles and obstacles. For vehicles, classification algorithms make it possible to distinguish between trucks, passenger cars, buses, motorcycles, etc., thus allowing predictions of their movements to be made. Traffic signs and traffic lights must also be reliably recognized and analyzed in terms of their significance for the vehicle. Creating models for such complex traffic situations is not only extremely difficult; the resulting algorithms also require substantial computing time in the vehicle. So-called machine learning methods are much more promising: with suitable training examples, networks can "learn" to understand what surrounding vehicles look like, for example, and so to recognize them in the real world. Such methods even make it possible to interpret inner-city scenarios in real time. We are testing these methods in both European and North American traffic situations.The more complex assistance functions become, the more important it will be to have a simple and unambiguous interface to the driver. This will take the fo...
