A group of the University of Tokyo and Nagoya University has succeeded in commercializing an "AI training system".
This group focused on the training instructors of driving schools as the target of the model driving model required for autonomous driving, and developed a driving model that rules the driving behavior.
In addition, by comparing the observation data of LiDAR (Light Detection and Ranging), which is a high-precision sensor specialized for autonomous driving, with a high-precision map, automatic driving software that can perform vehicle position estimation and obstacle detection in real time with high accuracy. Was introduced.We have also developed a technology to estimate the driver's face orientation using a machine learning model from images acquired by a camera installed in the car.
Using these as evaluation indexes, by evaluating the driving behavior that has been converted into numerical data using the driving model of the training instructor, the driving behavior such as how to move the vehicle before turning left and right, visual confirmation, shortcuts, large turns, etc. can be performed. An AI training system has been constructed that can be evaluated quantitatively with the same accuracy as a training instructor.In implementing this system in society, it also has a function to automatically control the brakes to avoid danger, as the training instructor does, for driving behavior that deviates from the threshold of the evaluation index and is judged to be dangerous.
Driving schools are said to have a problem of waiting for reservations for courses (2 to 3 months on average) due to the aging of training instructors and the shortage of human resources due to difficulty in hiring.By applying the system commercialized this time to driving schools, it is expected that it will be possible to solve these problems and to collect actual data that will contribute to the continuous development of autonomous driving technology.