The Uncertainty that the Autonomous Car Faces and Predictability Analysis for Evaluation

It is hard to keep the autonomous car safe because the environment in which the car runs has vast diversity and is quickly changing. As for system safety in the automobile field, ISO 26262 is one of the essential standards to achieve functional safety. For the car equipped with ADAS, recently the new guidelines are establishing; SOTIF (Safety Of The Intended Functionality) standards. Mainly it focuses on the limitation of sensors and human misuse. Because an ADAS car has many sensors to know the environment including other vehicles, pedestrians and so on. Also, the HMI has to bridge the gap between the human and machine-controlled area. We can extend SOTIF’s idea into the higher level of ADS (Automated Driving Vehicle), that is the autonomous car. There are many uncertainties when recognizing the environment. For example, an exact current location of the self-car, the other car’s information (speed/acceleration, azimuth), correct road condition, and so on. Especially it is hard to know the object behind the vehicle or buildings. The human driver can handle this situation with the ability of ‘prediction’. If we are near the school and can find the many parking cars along the road, we can easily estimate the possibility of jumping out of a child from a gap among street parking cars. This means we have the theme to think, other than the sensor performance or the excellent HMI, and also the future system must have the ability of prediction if running the environment having many uncertainties. Finally, we show the evaluation of the motion controller of the autonomous car from the viewpoint of ‘prediction field.