"Stanford University Dynamic Design Lab have completed a study that examines how human drivers respond when an autonomous driving system returns control of a car to them... The results of the study were published on December 6 in the first edition of the journal Science Robotics."
"The study found that the period of time known as “the handoff” — when the computer returns control of a car to a human driver — can be an especially risky period..."
They conclude with:
"The best way to protect ourselves from that period of risk is to eliminate the “hand off” period entirely by ceding total control of driving to computers as soon as possible."
Here is the related Teslarati article.
So they conclude that the best way to solve the handoff problem is to avoid handoffs by going to level 5 as soon as possible. "Doc, it hurts when I go like this." This seems like a non-answer to me. I agree that level 5 will eventually get here but there are a lot of corner cases that have to be worked out: construction, accidents, taking direction from flaggers/officers... These cases guarantee that there is some need for the next few year for a handoff to occur. Rather than the weak sauce of saying "don't do it", I'd like to see them researching better ways for the handoff to occur.
For example, they said that human steering response is tuned to the last speed at which we controlled the car, even if we are mentally aware of the speed change, our motor-response system needs time to adjust. Great. If you can define the problem, answers are the easy part. The steering response and stiffness can take this into consideration when handoffs occur.
With Autopilot (HW1.0 SW 8.0), you can press on the accelerator and it does not disengage TACC. I'd like to see this applied to the steering and brake too. (What? That's crazy!) Maybe, hear me out, then decide. There are a couple places on my commute that AP just cannot deal with (yet). One of them is a curve with an intersection so there are few lane lines. I know where these spots are and I take over as the car starts to encroach on the bike lane, before AP signals for help. However, just because I actively turn the wheel, does not mean that I wanted autosteer to disable. Of course, I want it to go the direction that I steer, but then afterward AP could still continue to follow the lane that it has now refound.
Tesla (rightfully so) wants us to keep our hands on the wheel when AP is on, but too often I find that this means I inadvertently turn off autosteer because I want to be slightly left or right of exact lane center (I have my reasons: pothole, rut, big truck, that guy on my left looks drunk...) and AP disagrees with me. I would like to see the AP system and the driver to work more in concert rather than as a tag-team. I would like to see the AP and driver act as a centaur chess team. This is another way to avoid the hand-off without removing the (currently required) human from the equation.
"The study found that the period of time known as “the handoff” — when the computer returns control of a car to a human driver — can be an especially risky period..."
They conclude with:
"The best way to protect ourselves from that period of risk is to eliminate the “hand off” period entirely by ceding total control of driving to computers as soon as possible."
Here is the related Teslarati article.
So they conclude that the best way to solve the handoff problem is to avoid handoffs by going to level 5 as soon as possible. "Doc, it hurts when I go like this." This seems like a non-answer to me. I agree that level 5 will eventually get here but there are a lot of corner cases that have to be worked out: construction, accidents, taking direction from flaggers/officers... These cases guarantee that there is some need for the next few year for a handoff to occur. Rather than the weak sauce of saying "don't do it", I'd like to see them researching better ways for the handoff to occur.
For example, they said that human steering response is tuned to the last speed at which we controlled the car, even if we are mentally aware of the speed change, our motor-response system needs time to adjust. Great. If you can define the problem, answers are the easy part. The steering response and stiffness can take this into consideration when handoffs occur.
With Autopilot (HW1.0 SW 8.0), you can press on the accelerator and it does not disengage TACC. I'd like to see this applied to the steering and brake too. (What? That's crazy!) Maybe, hear me out, then decide. There are a couple places on my commute that AP just cannot deal with (yet). One of them is a curve with an intersection so there are few lane lines. I know where these spots are and I take over as the car starts to encroach on the bike lane, before AP signals for help. However, just because I actively turn the wheel, does not mean that I wanted autosteer to disable. Of course, I want it to go the direction that I steer, but then afterward AP could still continue to follow the lane that it has now refound.
Tesla (rightfully so) wants us to keep our hands on the wheel when AP is on, but too often I find that this means I inadvertently turn off autosteer because I want to be slightly left or right of exact lane center (I have my reasons: pothole, rut, big truck, that guy on my left looks drunk...) and AP disagrees with me. I would like to see the AP system and the driver to work more in concert rather than as a tag-team. I would like to see the AP and driver act as a centaur chess team. This is another way to avoid the hand-off without removing the (currently required) human from the equation.
