A Warning to People Who Bike: Self-Driving Ubers and Right Hook Turns
sfbike.org
sfbike.org
Once an issue like this is found out, acknowledged as a problem, and a fix created, from then on it won't happen again (assuming no regressions). Gradually, the self-driving vehicles should become safer and safer; after several iterations, they should become safer than most human drivers.
That is how commercial aviation became one of the safest means of transport. Every significant incident was reviewed, direct and indirect contributing factors were determined, and new procedures and/or systems developed to reduce the chances of it happening again.
Unfortunately, it will take some time to get there. The initial implementations of self-driving vehicles are going to be less safe than the average human driver, and will remain so until enough experience is gathered. Not every accident is going to be of the "slow speed read-ended while waiting for the light" sort. Google has been getting away with it so far by putting a human on each vehicle, ready to take over on a hair trigger, while still trying to collect enough data to get over the safety cliff. I don't know how far they are, but by having the human assume so early, they are avoiding precisely the sort of difficult situations which could become important learning experiences.
I submit that most cars will be running very different versions of the code base, even if they are the same make, model, and year. In fact , because they are going to be in different jurisdictions, we can guarantee that this claim will be true.
I biked away with nothing but a large, painful bruise (and the car was dented), but this is definitely one of the more likely causes of collisions between bikes and vehicles.
The worst part was dealing with the insurance company. If this happens to you, get a lawyer and never talk to the insurance directly on the phone. Also get all the witnesses names and numbers if you can. The insurance adjuster claimed she couldn't get in touch with the cop (I couldn't either). If you're not in a car that has your own insurance backing it, they'll fuck you over.
I barely got enough money to cover my medical bills. I didn't get enough to cover all the bike damage, and had to buy used/mismatched parts to get it ridable again.
All cars must have liability insurance covering anyone you hit. You are not required to have coverage of your own vehicle (unless you lease or something).
It sounds like he is talking about the latter.
New Hampshire: requires a driver to provide proof that they can cover 25/50/25 in standard insurance speak (after being found at fault in an accident, I think).
Virginia: allows a driver to pay the state a $500/year fee instead.
Mississippi: allows a driver to post cash or bond.
I can offer the anecdote that you can get a drivers license in California by taking the exam in a rented car. Presumably you can borrow a registered, insured car (from a friend, say) in nearly every state?
This may be controversial, but I think bikers should be more cautious when approaching intersections where a right turn is possible from an automobile traveling at about the same speed immediately off their left shoulder.
Yes, I understand that cars are supposed to signal, and when they don't they are creating a dangerous situation. And they are supposed to merge into the bike lane before turning in order to block off cyclists from getting themselves into this exact situation. But they don't.
So cyclists should learn to look out for the signs of a possible right turn (drifting rightwards upon intersection approach is a pretty clear indicator) and be safe.
Don't approach an intersection at speed in the blind spot of an automobile that might turn right. Just slow down a little.
Depending on the exact position, that's a dead angle. The driver will NEVER see the bicycle.
The bicycle MUST stop if he knows the car is turning or he's going to guaranteed collision. Just sayin'
A couple of years back I wanted to walk over a crosswalk. The pedestrian traffic light was green. I looked left and right, saw a car approaching quickly, and stopped. The car ran the red light, drove over the crosswalk, and hit two crossing cars before crashing into a building on the other side of the road. The driver must have missed the red light.
Never assume that other people see you, never assume other drivers are as alert as you are, always be defensive, yield when you are unsure if the situation is safe, especially when you are vulnerable (pedestrian or biker vs. car).
Basically, don't speed through intersections on a bike, try to wear a helmet even if they are uncomfortable.
And if the cars are breaking the law, of course they should get a ticket like everyone else. But making a law that says self driving cars aren't allowed to break the law doesn't really make much sense. They're already not allowed to break the law. That's why it's called the law.
Drivers have personal liability. Self-driving cars have corporate liability. I'd expect a big judgment if someone is killed because of a known issue.
Yes, if a car is on the road and it's breaking the law, it should be removed from the road. But again, we already have laws that govern this. "But with a self driving car" isn't a special case. We've already written every law they need to follow.
People are killed from known issues with non-autonomous cars today, and companies are getting punished for them. There is already precedent. There are already laws. They exist.
Uber can absolutely fix known issues in its vehicles.
If you have more control, you have more liability.
Edit: Uber probably has product liability. Which is very different from liability for an employee's actions.
Perhaps it is the primary cause of collision between bikes and cars being driven by humans. Perhaps the primary cause of collision would be different for self driving cars.
Personally, beyond convention, I don't see why one of these turns would be inherently safer than the other. Either way, you still need to look behind you, in the lane to your right, in order to safely turn.
A self driving car has the advantage of being able to see in all directions at the same time (to the extent that it's sensor array allows). I don't think it's entirely clear that the safest way for a human to drive will always be the safest way for a machine to drive.
Whatever advantages you ascribe to it are the product of science fiction, not the evidence in front of us.
1: https://www.theguardian.com/technology/2016/dec/14/uber-self...
2: http://www.theverge.com/2016/12/16/13990318/uber-refusing-pe...
What's the downside?
As a cyclist, the difference is mostly one of expectations. If a car has slid into the bike lane prior to an intersection, I know that they will turn and that I should not try to pass. If the car does not do this, I may continue to bike next to them or even try to pass them in the intersection. This statement of intent is especially useful when the driver does not use their turn signal or uses it when they have already started their turn (which is often).
Also, it's the law [0] so it's not really up for debate.
[0] - http://codes.findlaw.com/ca/vehicle-code/veh-sect-21717.html
Having said that, of course, I am always on the lookout for cyclists.
I don't understand why TFA does not discuss signaling.
And apart from one unpleasant automobile-bike interaction this (and a helmet) has kept me alive.
What I do have a problem with is companies like Uber, Comma.ai, and to a certain extent Tesla who seem to not understand/appreciate how much work has to go into proving these things are safe before shipping them.
Fair? Probably not. But safety must be top priority here. We can't look like we're throwing this technology out there and hoping for the best.
And I don't think you can say Tesla's Autopilot 2.0 has been shown to be safer than humans.
Is there anyone who understands it better than the companies who are actually making them?
The engineers building the stuff probably yes but there are managers under high pressure to ship something, better be yesterday and costs should better be low. Chances are high that the second group decides to go for it in the very moment the first group says it's kind of ready.
If you have a company like Uber who generally push their products into the market with as much force as possible using hands full of VC money while ignoring all damage it may cause and always with full risk, this may even be worse. Also profit oriented companies not really caring about their impact to the general public (security,safety,health) is really not that uncommon.
Their goal is to be the first to market and swallow the relatively minor public image loss garnered from these accidents which, by their calculation, will be few. These losses will ultimately be offset by being a pioneer in the field and capturing a huge market share.
They might even intend to fix these problems, but they are by no means show stoppers. Companies do not have morals. They only have balance sheets and single digit number of lives barely even register.
"Uber car drifts in front of a school and kills 3 children." -- 5 more in the hospital. A witness captured the scene on his smartphone see video.
The first manufacturer to market half-assed will be the first to die.
Don't under estimate how much the public value their life and their limbs ;)
[1]: https://leginfo.legislature.ca.gov/faces/codes_displaySectio....
Uber's claim is that a permit is not required because a human is overseeing the system. But that's 100% disingenuous. The permitting program explicitly assumes a human is in charge. The point is to compel companies to report when the human has to disengage the system because it making an error or fails to intervene in time and causes an accident.[1] That information is crucial for assessing the maturity of autonomous driving systems.
Probably Uber's fail rate is super high, judging from multiple reports about them running red lights on day 1. They don't want to scare people off by reporting their early numbers.
[1] https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/testi...
C) Sometimes method "A" is best, because there's too much car traffic or too unsafe car traffic for the cyclist to take the lane (method "B"). In this case, it works wonderfully if the car yields to the cyclist (and any pedestrians). When cars don't yield to the cyclist/pedestrians going straight, and they should ALWAYS yield, you have problems.
D) Method "B" is ideal, except when the intersection sucks and doesn't detect the cyclist. There is an intersection I pass through everyday that is like this. If it's empty and I take the lane, it doesn't detect me. If it's completely empty (no cars in sight from any direction), I usually ride through even on a red. If there's cars, I ride to the curb, dismount, hit the crosswalk button, re-mount, and wait for green. The problem arises when after I do this, cars show up behind me to make a right turn and putting me in situation "A". Normally situation "C" plays out if the driver yields to me, but even so I always make eye contact with the right turners to make sure they're aware of me.
The best situation is B) if a car is already in the lane turning right or going straight as I approach the intersection. I can sit in the lane going straight and the light will go green because it detects the car.
tl;dr I wish intersection lanes detected bicyclists better. So much aged infrastructure.
The author admits there's no current practical problem: "Since yesterday, we have been told that “safety drivers” in these vehicles have been instructed to disengage from self-driving mode when approaching right turns on a street with a bike lane and that engineers are continuing to work on the problem"
Then he calls for a petition to ban the robot cars! I remember back when introducing daylight savings time was feared because farmers thought it would greatly upset their cows (in Alberta, this was.) As it happens I hate daylight savings time now, but not for that reason. Hmm, come to think of it, there is an argument here re unintended consequences - if we've spotted one already, there are probably more. I believe I've argued myself to a standstill. If I weren't so stubborn, I'd switch sides.
the above article lists one mitigation for the right hook: "Ride to the left" and, similiarly "Don't pass on the right".
Would this guidance (for bicyclists) be applicable in light of self-driving cars (in addition to human-driven cars)?
"Don't pass on the right" is only useful when traffic is going slower than bicycle speed (i.e. during rush hour when the streets are clogged, but not any other time).
I understand that the "right hook" kills a lot of cyclists, but I assume that's largely because of human driver blind-spots. Does the self-driving car have the same (or any) blind-spots?
This behavior should probably be fixed regardless, but I doubt these Uber cars are all rolling murder machines like the author seems to believe.
It seems like the government is happy to wait for the first serious accident that will no doubt immediately end whatever joke of engineering company finds its technology implicated.
I would not enter a reserved lane, but would by necessity cross over the intersection portion of that lane in order to take a right hand turn.
It sounds like San Francisco needs to modernize roadways they've improperly added 'bike lanes' to to include dedicated right turn lanes on the OUTSIDE of the 'bike lane'.
A better solution would be to isolate different types of traffic to dedicated pathways that are not directly adjacent.
Safety could be vastly improved if only there were a common standard for all drivers, pedestrians, and other users of the road.
The only problems are ( https://xkcd.com/927/ Standards ) and getting everyplace to adopt it.
Maybe we can get Google's Spinoff, Tesla, and the other self driving companies to just define one and push that in to law?
The city is starting to do this in quite a few spots.
The standard marking at the moment is to put a solid white line along the protected stretch of the bike lane, which becomes a dotted white line about 50 feet from an intersection. That's supposed to indicate to cars that they should merge if they want to turn right. In practice cars don't seem to realize this often.
Also, over years of consistent bullying from cyclists I honestly think drivers are in many cases afraid to enter the bike lane at all, even when they should do it to safely merge for a right-hand turn.
Anyway the correct answer is to do the full lane crossover as you suggest, but it's awkward on single-lane roads.
How on earth do cyclists "bully" drivers? Drivers aren't going to die if cyclists hit them.
I've seen a ton of hand gestures, cyclists physically hitting cars with their fists (or open-palm slaps), lots of yelling, and on one occasion I saw a cyclist stop in front of a car, dismount, and walk around to the driver's side to demand he roll the window down in order to engage in a verbal confrontation.
When you ride your bike to and from work every day for 7 years you see a lot of stuff. Believe me, there are a lot of self-entitled cyclist bullies out there making things less safe for the rest of us.
How on earth do cyclists "bully" drivers?
Well, there's this:http://sanfrancisco.cbslocal.com/2015/08/31/video-shows-crit...