The Harmonic Engineering Laboratory, Graduate School of Information Science, Hokkaido University and KDDI Research Institute, Inc. have succeeded in developing artificial intelligence (AI) technology that realizes "cars that meet each other".
A fully autonomous vehicle that incorporates autonomous driving technology, which is said to begin to spread around 2020.While the development of autonomous driving is underway, it is not good at coordinating with other vehicles such as "merging" and "passing", and in order to solve this, deep learning was used. Control by AI has become an issue.For example, driving control of a merging scene is realized by control elements such as deceleration, stopping, starting while checking the surroundings, and following the merging destination vehicle, and we are trying to make AI learn this series of control elements. In that case, if the follow-up to the merging destination vehicle is improved, a phenomenon of over-learning such as deterioration of deceleration / stop operation may occur.It can be said that this is because the driving scenes that AI should learn are not systematically organized.
Therefore, in order to improve learning efficiency in such complicated driving scenes, KDDI R & D Laboratories has extracted the target driving scenes and defined the requirements for ideal driving control including cooperation with other vehicles. In charge, Hokkaido University develops behavioral components (modularization), deep learning functions for each module, and AI technology that connects these modules.We confirmed the effectiveness of AI technology by mounting them on a radio-controlled car, which is about one-twelfth of the actual car, and learning the complicated movements at the time of merging.
In the future, by applying this result, we aim to develop "Yuzuri-ai" driving technology that is required in various driving scenes such as passing on narrow roads and turning right at intersections where straight ahead is prioritized.
