Fully automated driving anywhere is going to require special infrastructure, even with “AI” there is so much variety in road design, surfaces, markings, signs, etc. that even human drivers fail to navigate with alarming frequency (nearly 80K collisions in 20210). On top of that there needs to be laws put in place defining the minimum standard for a fully automated driving system, in the same way that driver testing sets the minimum performance of a human driver.
Industrial equipment, for example, has to have a safety evaluation that determines every way the machine could pose a risk to operators and how those risks can limited or eliminated. I don’t think it will be possible to prove how a fully automated system, based on how they work now, will respond to every possible situation making it inherently unsuitable for use on general purpose public roads.
Having a national (or preferably international) standard for how automated driving systems work and interact with each other would be a good first step.
I like that you pointed out industrial equipment. Those evaluations are usually pretty intense, and that’s within a very controlled environment.
Rule #1 of automation in my experience is: limit the scope of what you’re automating. Control as many variables as you can, make the requirements very specific. Every feature or situation or capability you add increases the difficulty exponentially.
Self-driving vehicles on the open road in the real world is an inherently unbound problem. The scope is nearly limitless. Good luck.
Fully automated driving anywhere is going to require special infrastructure, even with “AI” there is so much variety in road design, surfaces, markings, signs, etc. that even human drivers fail to navigate with alarming frequency (nearly 80K collisions in 20210). On top of that there needs to be laws put in place defining the minimum standard for a fully automated driving system, in the same way that driver testing sets the minimum performance of a human driver.
Industrial equipment, for example, has to have a safety evaluation that determines every way the machine could pose a risk to operators and how those risks can limited or eliminated. I don’t think it will be possible to prove how a fully automated system, based on how they work now, will respond to every possible situation making it inherently unsuitable for use on general purpose public roads.
Having a national (or preferably international) standard for how automated driving systems work and interact with each other would be a good first step.
0 https://tc.canada.ca/en/road-transportation/statistics-data/canadian-motor-vehicle-traffic-collision-statistics-2021
I like that you pointed out industrial equipment. Those evaluations are usually pretty intense, and that’s within a very controlled environment.
Rule #1 of automation in my experience is: limit the scope of what you’re automating. Control as many variables as you can, make the requirements very specific. Every feature or situation or capability you add increases the difficulty exponentially.
Self-driving vehicles on the open road in the real world is an inherently unbound problem. The scope is nearly limitless. Good luck.