SF Muni bus pauses for Waymo stuck in the street
San Francisco transit agency Muni has asked its drivers to carefully record the disruptions caused by self-driving Cruise and Waymo robotaxis operating on the streets of San Francisco. We have seen a number of reports online, in regular and social media, of blockages and incidents, including the recent accident between a Cruise and a bus. As a result, many wrote as if there was a serious problem. But anecdotes are not the same as data.
wired charger recently dug into Muni records and videos. Their title says the opposite of what the recovered data says:
“Agency logs show 12 ‘driverless’ reports from September 2022 to March 8, 2023, … Overall, the incidents resulted in at least 83 minutes of direct delays for Muni drivers, records show .”
As previously reported in their letter to the PUC of California, “Self-driving cars in San Francisco made 92 unscheduled stops between May and December 2022, 88% of them on streets with transit service, according to the San Francisco transit authority. city, who collected the data on social networks. reports, calls to 911and other sources, as companies are not required to report all outages. »
12 reports in just over 6 months
Considering the volume of traffic, having just 12 incidents in 6 months is a surprisingly good record, much better than many have led to believe from anecdotal reports. Data is not provided on the number of such events caused by human drivers, double parking and car breakdowns, but it seems likely that self-driving vehicles cause comparatively insignificant problems, as one might expect. expect it, because while they’ve probably covered a million miles in that time. between the two companies, humans drive a lot more.
The question is often asked to what extent robotaxis errors and problems should be tolerated during its development. It is to be expected that pilots and early deployments will lead to more errors and disruptions, and the only way to truly learn and troubleshoot is to be on the roads, taking the risk of such problems. Safety issues are much more of a concern, especially personal injury accidents. If people get hurt, there’s a stronger cause to measure that and make sure the risk stays within acceptable limits, like the risk we accept having human drivers on the roads.
Because robotaxis offer the promise – in fact, literally every company promises – of being safer in the long run, greatly reducing road hazards once at scale, it makes sense to tolerate some traffic disruptions in order to work for this remarkable goal. As long as humans drive most of the car, there will always be carnage on our streets, and the ability to reduce this is expensive, both in terms of money and traffic disruption.
The company already makes this decision when allowing learner drivers to take to the road. Students and freshly fired drivers are more dangerous and disrupt traffic more. We accept them because it’s the only way to get them to learn, improve and become safer, better and more mature drivers.
With robots, the profit is much stronger. Leaving a teenage student on the road helps make that driver better. The entire robot fleet is upgraded from everything learned from a single robot. If a robotaxi blocks a street or has a fender, the entire fleet will no longer make this error.
With a view to reducing risks, accidents and traffic disruptions, it would be foolish in the extreme to slow down the deployment of this technology because of small problems during its pilot phases. We would like to steal our future – and literally rob some people of their future – to avoid minor problems today.
Wired suggests there will likely be unreported incidents, which is certainly true, but these numbers are so low they could be significantly increased without reaching a threshold of concern.
The math is surprisingly clear. Although the robots are young, their deployments are small and their disadvantages inherently limited because of this. As they mature, they are deployed in large numbers and provide vast benefits. The ratio of future benefits to the present problem is on the order of thousands or even millions to one. Hundreds of less mature pilot cars are rolling out today, which will put millions of mature cars on the roads earlier tomorrow. In the absence of injuries, it seems odd to even debate it.
The main question is, how good must a company be before it can launch or expand a pilot project? It is clear that the vehicles could be so bad that their current problems would require a delay. Thanks to SF Muni’s study of the situation, showing only a few minutes of transit delay over 6 months, it’s very clear that Cruise and Waymo, at least, are well past this point. That Muni wanted to delay the expansion of its pilots is surprising.
Do better?
That’s not to say Cruise and Waymo shouldn’t do better. Cruise has had a lot more reports than Waymo and needs more improvements. Cruise has also had more crashes, including at least one with injuries, one with wires, and the one from 2 weeks ago with a bus.
A particular problem has been a conservative approach to helping cars when they run into trouble. Companies sometimes resort to sending a “backup” driver to physically get to the car and drive it manually to fix the problem. It may take 10-25 minutes, which is not a good situation. The plan for both companies is that they are able to resolve such a situation with remote operators who give guidance to the car over data networks. They relied on the rescue drivers enough to cause disruption and resentment. Cruise claimed he did it because he always wanted to make the safest choice, and a rescue driver is the safe and secure choice in a complex situation, but it’s the slowest.
Companies need to improve in this equation. The two decided they wouldn’t use “remote driving” where the remote operator has a console with a steering wheel and pedals and drives the car like a radio controlled car. It depends on very good data connections and always comes with risks that a back-up driver doesn’t have. Nevertheless, it has the advantage of something that can be done without delay. Businesses need to think about it or become more reliable with their remote support. Even though these incidents are, as Muni reports, extremely rare, they still attract a lot of attention which is amplified and reduces public confidence.