Yeah full autonomy that matches or exceeds human driving is a loooong way away for sure. Closing the gap is not just about sensing the environment, but also understanding it. That's (probably) a Hard AI problem.
Why is no one looking at retrofitting existing roads with markers that the car can use to follow? I've read suggestions of using magnetic markers which would have the added benefit of working through snow.
Roads get new lines marked all the time, seems like embedding a magnetic marker at the same time would be quite feasible. This would basically be a tram line on every street.
Who cares if the car needs to hand off to the human in the event of road works or missing markers? Those are fairly rare events on my commute at least.
Winter driving in cold regions is hard. Knowing where the middle of the road is and the car's position relative to it is only the very start of the problem.
In regions that have roads not cleared of snow for multiple days the lane positions change. The evolution of the position of the lanes is not strongly defined by where the previous lines were. The obvious spot to drive for a human would contradict the magnetic, ground penetrating radar database, or whatever marker is used.
Markers of any type would be useful for awareness of car self position but not sufficient for what the lane has become. Even as a human you have to guess sometimes because it's not possible to sense it. You just have to know what humans would do. Machines aren't great at heuteristics and guessing what humans might do.
> In regions that have roads not cleared of snow for multiple days the lane positions change.
Or even in the same day. I've seen major freeways shrink from three to two lanes after heavy snowsqualls blanketed the road completely in less than an hour. Yet, traffic still flowed smoothly and quickly.
Lane following isn't the hardest problem to solve. Current autonomous vehicle technology can follow lanes pretty well. The problem is with avoiding collisions while doing so, like if a pedestrian steps into the lane.
And if the vehicle discovers a magnetic marker is missing, by that point it's too late to hand off to a human. The human might not be paying attention. The lag time is too long for safety.
Why is no one looking at retrofitting existing roads with markers that the car can use to follow? I've read suggestions of using magnetic markers which would have the added benefit of working through snow.
Roads get new lines marked all the time, seems like embedding a magnetic marker at the same time would be quite feasible. This would basically be a tram line on every street.
Who cares if the car needs to hand off to the human in the event of road works or missing markers? Those are fairly rare events on my commute at least.