Re egress: There’s a big difference between emergency egress mechanisms and a called for emergency / voluntary stop — say, if you are feeling sick and want a temporary pullover, there’re commonly planned maneuvers for that in the design vs. if there’s smoke in the cabin then ofcourse a smart AV system will detect such degradation and pullover immediately. Now, you cannot cover for every single edge case — but you can absolutely develop smart, reliable, trustworthy conops to handle majority of the cases. AV tech companies partnering with ride-sharing are conscious of these design decisions as it has direct impact on the UX, and potentially impacting the ride share market value.
Re supervision: For handling true driverless operations, one must run a streamlined operational control that includes fleet management and in field support. Fleet management includes DMS like solutions to monitor packages of any sorts, and traditional CV like tech to identify malicious vulnerabilities. In field support can tackle any problems like someone sitting on top of the car hood at a traffic light.
Bottomline: Human machine interaction is a very very important part of mainstreaming the AVs but isn’t this a development in itself? We’re gravitating (even though very slightly one may argue) to talk about Ops from the traditional AV challenges - which ofcourse remains extremely important. I remain hopeful.
Thanks for the thoughts that add to the discussion! I really appreciate your point that there are levels of severity involved in pullover requests. Hazard and risk analysis should be performed for these just as is performed (or at least supposed to be performed) for equipment failures. The fact we are seeing news stories tells us this part isn't fully baked yet. That is no surprise. They should have a robust safety over-ride plan in place because these events were inevitable (several types predicted in UL 4600 for example), but pretty clearly they don't. This is another sign of them making it up as they go along and waiting for problems to find them rather than robust safety engineering.
"if there’s smoke in the cabin then of course a smart AV system will detect such degradation and pullover immediately" => I would not assume every company gets this right anytime soon.
" Now, you cannot cover for every single edge case" => Yes, my point EXACTLY. A human taxi driver can cover every single edge case with remarkably high accuracy (not perfect, but pretty good). The AV companies are still working on driving edge cases. Emergency edge cases is a whole other level.
Two related issues also need attention from regulators and developers.
First, if AVs do not allow unrestricted elective egress, are they committing false imprisonment? Clearly, a human taxi driver who traps a passenger against their will is breaking the law. What are the safeguards against and consequences for AVs that do the same?
Second, AVs without human supervision are ideal stealthy bomb delivery vehicles. The public is used to seeing robotaxis operating with no apparent human occupants. No one is checking packages they may be carrying. In fact, the business case for commercial parcel delivery such as Nuro and Uber Eats depend on humans not checking the package contents. What safeguards against bomb delivery are included in AVs with unlimited access to any address that protect the public? Ignoring this hazard, however unlikely, doesn’t make it disappear.
Even if AV developers one day solve the ‘safe driving’ issues, AVs will still be a long way away from safe.
Re egress: There’s a big difference between emergency egress mechanisms and a called for emergency / voluntary stop — say, if you are feeling sick and want a temporary pullover, there’re commonly planned maneuvers for that in the design vs. if there’s smoke in the cabin then ofcourse a smart AV system will detect such degradation and pullover immediately. Now, you cannot cover for every single edge case — but you can absolutely develop smart, reliable, trustworthy conops to handle majority of the cases. AV tech companies partnering with ride-sharing are conscious of these design decisions as it has direct impact on the UX, and potentially impacting the ride share market value.
Re supervision: For handling true driverless operations, one must run a streamlined operational control that includes fleet management and in field support. Fleet management includes DMS like solutions to monitor packages of any sorts, and traditional CV like tech to identify malicious vulnerabilities. In field support can tackle any problems like someone sitting on top of the car hood at a traffic light.
Bottomline: Human machine interaction is a very very important part of mainstreaming the AVs but isn’t this a development in itself? We’re gravitating (even though very slightly one may argue) to talk about Ops from the traditional AV challenges - which ofcourse remains extremely important. I remain hopeful.
Thanks for the thoughts that add to the discussion! I really appreciate your point that there are levels of severity involved in pullover requests. Hazard and risk analysis should be performed for these just as is performed (or at least supposed to be performed) for equipment failures. The fact we are seeing news stories tells us this part isn't fully baked yet. That is no surprise. They should have a robust safety over-ride plan in place because these events were inevitable (several types predicted in UL 4600 for example), but pretty clearly they don't. This is another sign of them making it up as they go along and waiting for problems to find them rather than robust safety engineering.
"if there’s smoke in the cabin then of course a smart AV system will detect such degradation and pullover immediately" => I would not assume every company gets this right anytime soon.
" Now, you cannot cover for every single edge case" => Yes, my point EXACTLY. A human taxi driver can cover every single edge case with remarkably high accuracy (not perfect, but pretty good). The AV companies are still working on driving edge cases. Emergency edge cases is a whole other level.
Two related issues also need attention from regulators and developers.
First, if AVs do not allow unrestricted elective egress, are they committing false imprisonment? Clearly, a human taxi driver who traps a passenger against their will is breaking the law. What are the safeguards against and consequences for AVs that do the same?
Second, AVs without human supervision are ideal stealthy bomb delivery vehicles. The public is used to seeing robotaxis operating with no apparent human occupants. No one is checking packages they may be carrying. In fact, the business case for commercial parcel delivery such as Nuro and Uber Eats depend on humans not checking the package contents. What safeguards against bomb delivery are included in AVs with unlimited access to any address that protect the public? Ignoring this hazard, however unlikely, doesn’t make it disappear.
Even if AV developers one day solve the ‘safe driving’ issues, AVs will still be a long way away from safe.