> When they became fully visible, our vehicle applied heavy braking but was not able to avoid the collision.
The "fully visible" part made me thinks humans can identify a bicycle and rider with just partial visibility, which means they would stop faster than the car.
Last year I caught this footage of a driver that hit a cyclist and they were fully visible. The police determined the driver technically wasn’t at fault because they had the right of way but humans drive incredibly unsafe.
1) cyclist assumed they were safe to cross as there was a pedestrian
2) the driver didn’t slow down for the pedestrian in the road at all
3) driver was probably speeding
4) driver didn’t brake until very late, probably distracted
5) cyclist was 15 years old so probably not using enough caution in general
I honestly think a self driving car would have been going slow for the crossing pedestrian and this collision would have never happened (but I guess we can all just guess).
The question is, what’s better, occasional software glitches or human stupidity? I guess time and data will tell.
Some states have a half rule. Or a rule that as long as the pedestrian hasn’t passed your car yet, you need to stop. I’m not even clear what my state (Washington) follows ATM:
> (1) The operator of an approaching vehicle shall stop and remain stopped to allow a pedestrian, bicycle, or personal delivery device to cross the roadway within an unmarked or marked crosswalk when the pedestrian, bicycle, or personal delivery device is upon or within one lane of the half of the roadway upon which the vehicle is traveling or onto which it is turning. For purposes of this section “half of the roadway” means all traffic lanes carrying traffic in one direction of travel, and includes the entire width of a one-way roadway.
I think they are trying to make the law more reasonable because there were police officers writing tickets for absurd stings. A person would step into the road and everyone would get tickets in the intersection. WA law seems rather convoluted but basically gives some room to use the road when the pedestrian is still out there. Probably more for those, I’m turning right at a red light and ped crossed the road, it’s clear to go now but someone is still technically in the crosswalk.
IMHO it's a slippery slope and vision of other cars can be obstructed by cars driving around pedestrians.
In general, people shouldn’t be doing 40mph within 10 feet of a pedestrian in the road (like they were in this case).
Anyway, I record this intersection because I see all kinds of carnage… 10 collisions in the past year next to a park. Human drivers are the worst
Even if a human sees the bike slightly sooner, the reaction time is so much slower that it's either a wash or the self driving car is likely still faster. Median braking reaction time for humans is ~500ms[1].
I can't read that article, but I suspect they're not including all the required foot movement there.
This study [1] shows ~2 seconds from gas lift to max brake force, when using right foot.
Eliminating the foot movement is why I left foot brake. I feel like I have some superpower with how fast I can react, compared to others. After all, we were taught to brake with our right foot out of tradition, not data!
> The "fully visible" part made me thinks humans can identify a bicycle and rider with just partial visibility, which means they would stop faster than the car.
From my experience as a driver and a bicyclist, I guarantee you that is not true, most driver won’t see you unless you are >60% visible.
Based on what? Humans run in to cyclists too and although I have know idea about Waymo's tech, in a research setting computers are better than humans at figuring out what objects are based on visual data.
If computers are so good at this, why am I (a human) asked to identify bicycles and motorcycles to train these computers when confronted with a captcha? A human needs far less training and can identify a person on a bike without ever seeing an actual bike. They can also infer what is likely to happen based on a lot of context clues. Even if an event doesn’t seem likely, I can still prepare. For example, if I see kids playing basketball in a driveway, I’m going to slow down, incase the ball bounces into the street and a kid takes off after it without looking. I don’t need to wait for the kid to be in the street to start reacting and prepare for different possibilities.
> A computer doesn't need a multi million dollar marketing campaign convincing it not to text while driving or drive drunk.
A human doesn't get random bugs every week from OTA updates. The human mind is not a viable target for hackers. The risks of human drivers are individual. The risks of computer drivers are systemic and massive. One bad bug or hack could kill millions if self driving cars become ubiquitous.
>A human doesn't get random bugs every week from OTA updates.
What bugs are you referencing exactly? Humans have something called emotions that are far more variable than bugs in a safety critical system and you're exposed to thousands of different flavors every time you get on the road.
>The human mind is not a viable target for hackers.
But it is, and not even sophisticated ones at that. One lane change and a middle finger from a fellow driver can induce a murderous road rage.
>The risks of computer drivers are systemic and massive.
Hypothesis. Reality: 45,000 people die every year in the US from human drivers. That is systemic and massive.
You still don't get. All human problems are individual. Emotions in one person have zero impact on emotions in a driver two state away. Software problems are systemic. They change the entire system at once.
Your "just hypothesis" dismissal would carry a lot more weight if silicon valley hadn't been cranking out bug filled garbage for the last 20 years in their move fast and break things frenzy. Kyle Vogt was (in)famous for vaunting that culture at cruise so there's no argument AVs have somehow escaped the silicon valley rot. The tech industry need to stay far, far away from anything safety related
With the hacker comment, in fairness the weakest link in a lot of orgs is often the human one. Ignoring obvious stuff like phishing links, people can be disillusioned by their employer or their government through propaganda and other campaigns run by their adversaries. The westerners who supported ISIS, etc didn’t just do so in a vacuum out of the blue.
>The captchas are just solved problems being used to slow down bots and spammers these days.
If bot are solving these problems faster and more accurately than humans, it’s not really slowing down the bots and spammers anymore. That is also a solved problem.
Well, if it is solving CAPTCHA better and faster than humans, that is a useful signal. Maybe they monitor how often an IP address figures something out and if it gets too many right they +1 a suspicion counter. I dunno. Implementing the CAPTCHA solving bot also takes time and expertise; it'd filter out some low-ability spammers.
>if I see kids playing basketball in a driveway, I’m going to slow down, incase the ball bounces into the street and a kid takes off after it without looking
That's because you're a thoughtful, conscientious driver. A very large fraction of drivers are not, so society winds up with lots of dead kids. Humans are terrible drivers: some of them are OK, but too many are very bad, leading to catastrophic results, and because of this human drivers need to be eliminated. Alternatively, the society could become far stricter about which humans are allowed to drive, but that isn't feasible when the society has intentionally made it so that it's effectively impossible to survive without owning and driving a car.
The captcha example isn't exactly fair, the images shown can be newly generatrd, repeated, or used as control images, and even if they are unclassifiable then you're not seeing the successes.
Without knowing the whole system there's too much hidden bias to claim the computer is more of less accurate than a commuter.
No doubt all of your excuses are at least partially true, however, at this point, literally after almost two decades of billions of people training these things to see bikes, it still needs a lot more work.
How many years or decades did we spend answering them before it got to where it today?
I’m all for optimism, but these problems are clearly much harder than the futurists would like the public to believe. Elon said Teslas would be driving across the country to pick up their owners by 2018. 6 years later and what does it do, drive across a parking lot?
Trains are also a solved problem, and seem like a much better solution to the problem. With rails, a vast majority of the complexity around driving goes away. They also remove traffic instead of keeping it the same or making it worse. Less traffic would likely mean fewer accidents, even if humans are still the ones driving. And batching transit together, with dozens of people in one vehicle is going to be far more energy efficient than everyone running their own vehicle. It also solves the issue of drunk driving, distracted driving, and bad drivers in general, as everyone is just a passenger.
So much money, time, talent, and energy has gone into this self driving stuff, and it seems like a massive misallocation of resources. They’re solving the wrong problem.
Seeing a bike and inferring the vector of a bike are very different. A kid or tweaker on a bike needs more frequent updates than someone commuting or training.
The "fully visible" part made me thinks humans can identify a bicycle and rider with just partial visibility, which means they would stop faster than the car.