When snow or bugs shut down an entire robotaxi fleet

Even bicycles can suffer a shutdown

The goal of a successful robotaxi fleet is to offer a service that works as a "car replacement." You need to get people to give up car ownership to attain the brass ring. If they still own a car, they are motivated to use the expensive vehicle they already paid for rather than use your service. Most commonly in the early years, this will mean convincing people to switch from 2 cars in a family to 1, and use the service for all times 2 or more people need to travel.

If you have a great robotaxi service, with low costs, short wait times, and a wide service area, it should be easy to win such customers. You may even win them with special subscription plans which mirror the cost patterns of car ownership, but at a lower cost and the pleasure of not having to drive.

One thing that might scare them away would be the service becoming unavailable. Once people decide to depend on a service to commute or do other vital trips, they won't tolerate it being down with any frequency.

In particular, I want to consider these problem areas:

  1. The vehicles are not rated for certain weather conditions, like a certain amount of snow, and they all stop service when this happens
  2. Some major fault in the fleet management system makes it impossible to summon cars. (Airlines have shut down fairly often due to this.)
  3. A safety problem in the software is discovered that is so serious that the fleet manager decides to cease operations until it's fixed.

The other modes of transportation also fail. Personal cars break down and need to be in the shop. Taxi demand can spike. Uber and Lyft can have unaffordable "surge" prices. Transit operators go on strike (especially in Europe.) Transit lines shut down due to problems on the line. Highways get temporarily closed. And weather even gets so bad that most human drivers refuse to go out in it.

When this happens we have solutions. When the personal car fails, it's fairly easy as transit, Uber and other methods are still working. It's more challenging when a fleet or mode fails.


Most robocar developers today, in a "one problem at a time" approach, are putting minimal effort into snow driving. There's lots of places to get started without snow. Even so, there are some promising solutions in development to deal with the problem of not being able to see the road, including radar that reads the gravel underneath the pavement and 3-D and edge sensors that figure out where you are from the patterns of trees, poles and buildings.

Even so, humans are reckless in snow. We'll go out driving "because we have to" in snow we probably aren't safe to drive in. Commercial fleets will be forced to be more conservative on that account. So a day will come when your robotaxi service says, "We're not operating until the snow clears" but people are out driving regular cars. In this situation, transit, Uber/Lyft and carpooling may suffice depending on how bad the snow is. Beyond a certain limit, nobody will be angry at their robotaxi service.

Still, it is a different situation in towns with rare snow compared to towns that get dozens of snowfalls a year.

Software failures

Regardless of the weather, software is not perfect, so bugs may occur that shut down the fleet -- either directly, or because they compel the fleet operator to order the shutdown. Most frightening, perhaps, is the discovery of a safety flaw of serious proportions. Right now in the testing phase, it is quite common for the discovery of a problem, or worse an accident, to cause the grounding of the entire fleet until the problem is isolated. When Uber's vehicle killed Elaine Herzberg, their fleet was shut down for what is now 3/4 of a year. Once a fleet is widely deployed, accidents will happen, perhaps several times a day, and you can't shut down the fleet every time one happens. But some accidents will be serious enough, or signal some serious unknown problem, that the decision may be taken.

It's a tough call. When such an event happens, the vehicles haven't changed. They've been driving around for a long time, having an acceptable safety record, and the arrival of an incident doesn't change that. The problem is that from a legal standpoint, the problem is now something they know about. If a car causes a problem for an unexpected reason, that's one thing. If the car goes and does it again, after the operator knew the problem was there, that's another, at least in the legal system.

This challenge is worthy of another article, because after such incidents the team will need to find the problem, create a fix and deploy it. But a fix deployed in a rush can be a risk of its own. But not deploying the fix is a risk as well.

In the old world, when a safety flaw is found in a car, the maker issues a recall. They send letters to all car owners, warning them of the problem and telling them to come into the dealer for a fix. Once they are notified, if they keep driving the car without going for the fix, then the driver is assuming a lot of that risk. Not the case in a robocar. The order not to use the car can be instantaneous.

Unless they never find a problem so serious, or they never have a fleet management shutdown, it's going to happen.

Contingency plans

A smart robotaxi fleet operator will need contingency plans to maintain customer confidence in the event of shutdowns. These plans can involve

  • Paying for human driven service (Uber/Lyft/Taxi) for customers -- if such services are still operating in the robotaxi world.
  • A reciprocal agreement with competitors that they will handle the load in case of a software failure.
  • Having a fleet of "standby drivers" ready to be become taxi drivers for a day. They may even drive the robotaxis themselves, through use of a plug in steering wheel and pedals.
  • Recruiting, in advance, a large number of people willing to be carpool drivers during the outage, going only minimally out of their way.
  • Free transit tickets for all stranded customers, and possibly even deploying fleets of contracted vans and buses
  • In some cases, allowing customers with drivers licences and some practice to use plug-in controls to drive the vehicles themselves -- though this may not be a good idea in bad driving conditions.

As much as possible, the backup plan must deploy with few hiccups. If a service leaves customers in the lurch too often, stranded, they will possibly abandon the service.


Yes, this is certain to happen. To minimize risk, every city/metro/etc should strive to have ideally three, but minimally two robotaxi providers, so when one has to shut down, capacity is reduced but service is still available.

Indeed, a smart provider has a reciprocal agreement with its competitor, that they each will handle problems by the other. This does not help in snow, though.

The reciprocal agreement scheme also depends on that competitor having an improbably large surplus capacity.

Or that there be multiple competitors. Or that the competitors agree they will push people into carpooling as part of the deal. Their customers won't like that but they will like it when it goes the other way.

'We'll go out driving "because we have to" in snow we probably aren't safe to drive in.'

Let's address that snarky "because we have to" bit. How do we, in snow country, assess whether or not to get into, or stay in, the car and hit the road? I have a job where I can look out the window in the morning and decide to call my boss while making some hot cocoa. My boss, who has his own window, will understand. But suppose I were an ER physician. That changes the calculus. In really extreme weather the national guard will go to those guys' houses to pick them up and take them to the hospital. There is, however, a lot of ground between where I am making that cocoa and where you need a humvee to get through. In that middle ground, our ER doctor just gets in his car and drives carefully. But back to unimportant me. When might I drive in that snow anyway? If I have to pick up my kid, that changes the calculus. Pick her up from where? If it is a friend's house, then she might get an unplanned sleep-over. If it is from the library, then it has to be humvee-only levels of snow. How about if I am driving home when unexpected snow hits. It happens. The forecast says it will be 33 degrees and rain; then the temperature drops two degrees. This is one of the worst scenarios, because the roads will be untreated and the plows won't be deployed properly. The highway can become a complete mess. If I am among the mess, it may or may not be possible to pull off and hole up in a coffee shop. If not (and it probably isn't) then the choices are stop where I am and wait (for how long? Hours? Days?) or keep going if it is at all possible. Keeping going is by far the better choice. The last time this happened to me, I took two hours for what would ordinarily be a half hour drive.

What would a robocar do in these situations? Does the national guard have to come pick up that ER doctor in even moderate snow? Do the robocars in that mess of a highway shut down when a car with a human driver could keep going (if it weren't for the road being blocked by inactive robocars)?

Yes, there are idiots who go driving when they shouldn't, because they can't tell the difference between necessity and desire. But even for reasonable persons "have to" is a sliding scale dependent on specific circumstances. When a reasonable person has a legitimate need that a conventional car can meet and a robocar cannot, this is a problem for robocars.

Being a Canadian, I am pretty familiar with snow driving, and know I've taken some stupid risks doing it.
But the key factor is that driving in bad snow conditions means taking extra risk. People can decide they will take that risk because getting there justifies that risk. And society has come to accept that, though it punishes you, of course, if a bad things happen to somebody else because of it.

But it's much more difficult for companies to act that way. If a company says, it's important to keep the service up because our customers need it, even though driving is clearly much less safe" they are taking a much larger risk upon themselves, and on their whole system. We let individuals take risks we don't let companies take.

Also, I love how "a safety flaw of serious proportions" promptly gets euphemized down to an "incident" and those darned lawyers blamed for this being something to worry about.

A wide range of incidents could lead to a decision it's not safe to operate the vehicle. Obviously being at fault in hitting something or someone is the most extreme type of incident, but there could be equal concern for an incident where the car only escaped doing so by luck. (In fact, with humans, almost accidents happen all the time, but when bad luck comes, they turn into crashes.)

I believe that 99.9% of all snow/ice related accidents are the result of something robocars can be immune to - going too fast. If the speed limit is 55 and it seems to normal humans safe to go 40 but really it's only safe to go 20, how fast will a robocar go? If it's designed right, 20. The impatience of going much slower than normal is tempered considerably by all of your time as a rider being free to go about your business. Hopefully the robocars frustrate and annoy human drivers into driving sensible speeds in rough weather.

Or it might know it has better reflexes and driving ability. Of course, so do humans now, thanks to ESC and ALB it's hard to spin out any more, though you can certainly skid to some degree.

A robocar with encoders on all 4 wheels will know precisely when the wheels are slipping and get a good sense for traction on the road. It might even publish that info up to the cloud for other cars to receive when they drive over that piece of road.

Add new comment