This talk addresses both the socioeconomic and technical issues which are behind the development of the next generation of cars. The talk will focus on a key issue: How to develop a robust and efficient embedded perception system, having the capability (1) to fuse multi-sensors data while reasoning about uncertainty, (2) to represents and interpret in real-time the surrounding dynamic environment, (3) to make use of contextual information and prior knowledge for predicting world changes and evaluating the risk of future collisions, and (4) to give insight for appropriate driving decisions. It will be shown that Bayesian approaches are mandatory for developing such technologies and for obtaining the required robustness in presence of uncertainty and complex traffic situations. The talk will be illustrated by some results obtained in the scope of several projects with Toyota, Renault or IRT Nanoelec.