What is it about?
Our automatic vehicles have learned to safelly navigate a crossing, keeping track of the other vehicles and pedestrians. Yet, when there are traffic lights the behavior of the other traffic changes, and the automatic vehicle should also change its behavior. For instance, it should slow down and stop precisely before the stop-line, waiting for a green light. That seems obvious, but from far away the traffic light is just a few very important pixels in an overload of other colors. Here the automatic vehicle is trained to take care of this few pixels and adjust its behavior accordingly.
Featured Image
Photo by Yena Kwon on Unsplash
Why is it important?
As long as not all traffic lights send out wireless signals (and even when they would do, a backup system is still needed), automatic vehicles should use their camera's to recognize traffic lights and adjust their behavior accordingly.
Perspectives
At the beginning of this study I was afraid that the few coloured pixels would not be enough to adjust the behavior. I was wrong. What was actually the most difficult is to fine-tune the approach behavior, to slow down smoothly and stop at time near the stop-line. That was actually harder to learn in end-to-end fashion.
Dr. Arnoud Visser
Universiteit van Amsterdam
Read the Original
This page is a summary of: Training Traffic Light Behavior with End-to-End Learning, January 2023, Springer Science + Business Media,
DOI: 10.1007/978-3-031-22216-0_50.
You can read the full text:
Contributors
The following have contributed to this page
