Researchers at Cornell University have found a way to give autonomous driving a memory.
the AI in autonomous driving, real traffic data is used trained, but she has no memory that keeps the data. You only learn from it. But that could change now.
The vehicles always see the roads they drive on for the first time. You have no memory of the past. Now researchers of Cornell University Developed methods to make the AI remember. These memories are intended to carry experiences which will be of benefit in future navigation. This is especially true for adverse weather conditions, if the sensors are not working so reliably.
The research team led by Carlos Diaz-Ruiz made one record, which consists of repeated segments. The lidar data from a year and a half and 15 kilometers in length around Ithaca imply motorway routes, city driving and that in rain, snow and sunshine. In this way, the AI should be able to remember the course and the objects, even if there is snow on the road or rain impairs the sensor view.
The data of each LiDAR point is stored in the local SQuaSH database and continuously updated. Through the repetitions, the system also learns which objects are dynamic and which are stationary.