The challenge of roadside mirrors on places with low visibility

[Senter]

In the area I live there are are often some situations (e.g. intersections or merges) where roadside mirrors are mounted to help drivers cope with the low visibility and high risk. Creeping onto the road in those places would be incredibly dangerous as oncoming traffic is extremely fast and sometimes comes from behind a curve and a rock.

There are people way smarter than me here… I was wondering whether an AI can learn how to cope with such a situation at all? I guess radar and lidar cannot really do that and visual approaches would struggle with the fact that objects coming from the left can be identified by looking forward…

 

[diplomat33]

In the area I live there are are often some situations (e.g. intersections or merges) where roadside mirrors are mounted to help drivers cope with the low visibility and high risk. Creeping onto the road in those places would be incredibly dangerous as oncoming traffic is extremely fast and sometimes comes from behind a curve and a rock.

There are people way smarter than me here… I was wondering whether an AI can learn how to cope with such a situation at all? I guess radar and lidar cannot really do that and visual approaches would struggle with the fact that objects coming from the left can be identified by looking forward…

In theory, it should be possible to train camera vision to correctly understand the roadside mirrors. You would need to train the camera vision to recognize that it is a roadside mirror and then detect the objects in the mirror and correctly "project" the objects in the right direction. It would probably be very hard to do reliably. It would be best to train the car to slow down correctly. If the car had perimeter sensors, it could help the car peek around sooner. AVs like Waymo have perimeter sensors for situations like this.