During a webinar today one of the attendees said that Chinese robotaxi companies are now upward of 10 vehicles per remote supervisor. It is difficult to believe they can maintain situational awareness of that many vehicles, and eventually there will be a situation in which multiple of one supervisor's vehicles needs intervention at the same time. I suppose the argument is that soon they will not need any supervision so this is a transient phase. It is inevitable that there will be business cost pressure to go this way, at least until the first really bad mishap. I suspect there might be a bit of wishful thinking involved in this situation.
Good point about remote drivers. Even driverless subway trains are not fully driverless, and there are some people watching permanently on remote through screens.
Another systemic problem might be that there is no digital pucker factor functionality in neural networks. Humans, at least in my personal experience, have serious pucker factors and adrenal pumps that kick in when something fails or the driver senses an approaching dangerous situation. Increased alertness and situational awareness rapidly follow. That’s because I (and likely you) care deeply about the outcome and likely consequences of the rapidly deteriorating driving safety margins and adjust not only due to what we see but also to what we expect to see, hear, or feel, and project that onto other vehicles or threatened road users. Sure, AVs don’t get drunk, but they really don’t care.
During a webinar today one of the attendees said that Chinese robotaxi companies are now upward of 10 vehicles per remote supervisor. It is difficult to believe they can maintain situational awareness of that many vehicles, and eventually there will be a situation in which multiple of one supervisor's vehicles needs intervention at the same time. I suppose the argument is that soon they will not need any supervision so this is a transient phase. It is inevitable that there will be business cost pressure to go this way, at least until the first really bad mishap. I suspect there might be a bit of wishful thinking involved in this situation.
Great overview thank you.
Good point about remote drivers. Even driverless subway trains are not fully driverless, and there are some people watching permanently on remote through screens.
Thanks.
Another systemic problem might be that there is no digital pucker factor functionality in neural networks. Humans, at least in my personal experience, have serious pucker factors and adrenal pumps that kick in when something fails or the driver senses an approaching dangerous situation. Increased alertness and situational awareness rapidly follow. That’s because I (and likely you) care deeply about the outcome and likely consequences of the rapidly deteriorating driving safety margins and adjust not only due to what we see but also to what we expect to see, hear, or feel, and project that onto other vehicles or threatened road users. Sure, AVs don’t get drunk, but they really don’t care.
So, does anyone have a good sense as to how much typical "teleoperation" is costing robotaxi companies?