Autonomous car : “We must never let the algorithm decide in the last resort “

Forum. Ten years ago, there was the first autonomous car : Google car. Still ten more years and, we are told to all corners of the World of the automobile, our streets and highways will be crossed by the new gear.

Summary : an autonomous car is a vehicle equipped with measuring instruments to acquire a knowledge of spatial and temporal environmental : half-a-dozen cameras, radars and lasers to determine position and speed of moving objects around, such as the five senses of an animal. These sensors are connected to a computer in which the software, from these data, determine the action to be taken : brake, turn, avoid, etc

To continue the analogy with biological, this software is essentially based on artificial neural networks. It is in summary a great memory organized in layers to which one has learned the situations in which the vehicle will be facing in the future as well as the actions that flow from it. A little like, to act, our animal has an innate knowledge and experience. Here, it is obviously a question of acquired. And that is probably the problem.

Uncertainty associated

The functioning is the following : at each instant, the program compares the present situation with those learned, selects the closest, and acts according to what he has been prescribed : by imitation, so, in a kind of conditioned reflex. At the bottom, the weather forecast has long worked in this way : the current situation is compared with a very long history in order to deduce what is likely to happen.

Note that it is not a question of mechanical methods, deterministic, in the sense that the calculations determine a trajectory as it is predicted orbits. It is not to understand, to predict, as science does, but to accumulate enough experiences to predict the future by analogy. From this point…

Like this post? Please share to your friends:
Leave a Reply