Uber dares to return, Senate pushes law, and query about recharging scooters

Some news items, and then some analysis of the energy needed to reposition and charge all the dockless scooters from Lime and Bird.

Uber coming back?

The New York Times reports that Uber will begin autonomous testing again but just on a one mile fixed route between their facilities and at 25mph or less. They know that if they have even the slightest incident they are toast, so they need to keep it to this fairly low utility level of testing. But they do seem to want to come back. Even so, the Times story includes internal sources saying they have problems and shouldn't do even this. As their earlier safety documents show, they plan to do it with the obvious things fixed.

But can anybody operate in an environment where no mistake can be forgiven? Uber used up all the slack the public might have offered them.

Senate tries self-driving bill

Earlier, the house passed a self-driving bill unanimously, but the Senate version got stalled. It offers an clause forcing disputes to arbitration which has bothered several groups, not without reason. I sort of agree with the concept, but think it's a little early to implement it.

Perhaps the greatest remaining legal risk for robocars in the USA is the chance that they might have fewer accidents than humans, but each accident will cost many times more than it does when a negligent human causes it. This is bound to happen the way the courts work. Each accident will be different, and involve a lengthy trial with depositions of all the programmers, expensive expert witnesses and lawyers looking to make a score and a name for themselves with famous, deep pocketed defendants. Compare that to regular car accidents which are settled quickly by the insurance system and almost never go to court.

I fully expect the first self-driving accidents to be very expensive. I was wrong about that, because the first one killed a homeless woman who did not have legal right of way, about the most fortunate legal circumstances Uber could have had. But my argument still makes sense, and so we'll see these expensive cases -- which the likes of Waymo, GM and Uber can afford. But it can't stay that way forever. Eventually the costs have to come in line, and legislation might be needed to fix that. But we don't need to fix it beforehand, just be ready to do so.

Here is Wired's coverage of the Senate bill.

Measuring the efficiency of scooters

In Antwerp, a truck repositions a large number of docked shared bikes

I am planning an article about the energy efficiency of the newest form of computer-enabled transportation, the dockless scooters from companies like Lime and Bird. While already raising ire for other reasons, their energy efficiency is absolutely astounding. The typical scooter of this sort uses about 18 watt-hours per mile! To put that in context, a Tesla is about 250 per vehicle mile and the average light rail is 370 watt-hours/passenger mile. Yup, those guys on the scooters are using one twentieth of the energy of all those folks on the light rail, in aggregate. That's so incredibly good that the efforts to ban these scooters are ill-advised, it is better to work for solutions to make them better.

However, the number is misleading. The reason the light rail is so poor is because it carries a lot of empty seats. The real number that matters is how many full seats something carries and the true total energy. The scooters are almost that efficient in docked scooter systems, where they dock and recharge at fixed points, but they are popular because dockless ones are much more convenient.

Each night, however, Lime and Bird and the others send out large crews of gig workers to collect depleted and out-of-zone scooters and recharge them and reposition them. The workers get from $5 to $20 to do this to a scooter. I'm trying to figure out the energy required to do that, because I think it's quite possibly higher than the energy put into the battery to move the customers (about 320 watt-hours in a typical model.)

Some of these rechargers are super efficient. They collect scooters using other scooters. They have found (crazy) ways to stack 5 scooters onto one scooter, and they zip around depositing them. Some of them also pick up and redeploy scooters along their commute, thus using almost no additional energy over the travel they were already doing. (It's a little hard because you must collect after 9pm and deploy before 7am.)

On the other hand, many are going out in cars, vans and pickup trucks to collect the scooters. The efficient ones are able to collect a lot in a small area (in the dense urban zones) and do an efficient deployment trip. They might drive a total of 5 miles to process 10 scooters -- just 1/2 mile of driving per scooter. But I've also talked to chargers who, unaware of anything besides the cost of gasoline, will drive as much as 10 miles round-trip for a single scooter.

The scooter companies track the units with GPS, so they could calculate these numbers exactly. I have asked them all to help and gotten zero response. This suggests they don't like the number. (Or rather, they don't know it yet but are afraid of what it could be.)

A gasoline car doesn't use watt-hours. However, on a rough average, if you burn a gallon of gasoline in a generator you might get around 12,000 watt-hours. That's a pretty rough figure, If we go with that, the guy stacking scooters is spending just 10-20 watt-hours per scooter moving them around. The guy doing 10 scooters in 5 miles in a 25mpg car is using 240 watt-hours (of gasoline) per scooter. In a 15mpg truck it would be more like 400. The long-distance driver is using a whopping 4800 wh to recharge a 320 wh scooter. Let's hope there are not too many of him.

In addition, not all the scooters are empty. I don't know how empty a scooter has to be before they tag it for recharge. They do want most of them repositioned to "nests" (as Bird calls them) in the heavy traffic areas. So if a scooter only got half its battery used the ratio is worse.

In the end, only the companies know, but I am interested in input on back of the envelope calculations.

On the plus side, I think the companies could make this much more efficient. Right now, they let the chargers just wander around grabbing scooters they can find. People often arrive to collect scooters 20 seconds instead of somebody else -- that's wasteful. Some people collect them (against the rules) before 9pm and "hoard" them in their backyard to claim all the bounties.

Instead, the companies could allocate efficient groupings of scooters to a single contractor. They could generate a series of optimum routes with minimal travel, based on input from chargers on where they live and what their commutes are (though they will lie about that.) They can readily stop the hoarding and probably get the driving down below 1/4 mile per scooter. They could also reward collecting in efficient vehicles like electric cars and hybrids, though that's also harder to enforce.

At a rough guess though, I'm going to estimate that the recharge/redistribute today uses at least as much, perhaps twice as much energy as the riding of the scooters does. That sounds terrible, but don't forget -- the scooter is super, super efficient on its own, so making it 2x or 3x less efficient it still wins by a large margin over anything. Anything except better managed scooters and e-bikes.

As noted, docked vehicles (with charging at dock) don't need to be moved for charging. They do often need repositioning, but with docks, that's usually a small truck with a trailer that efficiently moves scores of bicycles or even a hundred scooters at a time. In the photo above, I show a truck moving the bikes in Antwerp.

This is part one -- I have more analysis of the scooters coming up.

Have any readers taken a gig moving scooters? What numbers do you think make sense?


"on a one mile fixed route between their facilities and at 25mph or less"

That's just inviting trolls/protesters.

The thing that most people don't understand about traffic accident cases is that in most of them, liability isn't a major question. There are two halves to the case: liability (roughly, whose fault was it?) and damages. Damages in turn consist of economic damages (roughly, anything with a number attached: medical bills, car repair, lost wages) and non-economic damages (often called "pain and suffering"). In most car accident cases, damages are the real dispute.

Liability is usually, though not always, straightforward. The most straightforward is the common case of someone getting rear-ended while stopped at an intersection or in traffic. There isn't really anything to argue about there. When we say that these cases almost never go to court, this vastly overstates the case. Most don't go to trial, but it isn't astonishing when they do. Often when they do, it is those instances where liability is in dispute. Most cases settle because an experienced lawyer can make an educated guess as to a likely judgment for a given fact set in a given jurisdiction. If both sides have experienced lawyers, they will see eye to eye on the likely outcome, so why go to the expense of schlepping to the courthouse?

The real question with self-driving is what the plaintiff will have to prove. If this is treated as a product liability case, then the plaintiff has to prove the product was defective. This involves lots of demands for documents from the defendant, and very expensive experts to testify. If it is treated like any other traffic accident case, that is as negligence, then the fact that one (or both) of the vehicles was self-driving is essentially irrelevant, except perhaps that there will be better information about what exactly happened. But it will revert to the usual scenario where the real argument is over damages. You aren't hurt more because it was a self-driving car that hit you.

If you want a prediction over the early legal battles, it will be over this question. Impress people at cocktail parties by encanting "Res ipsa loquitur." Or to sound really in the know, just go with "Res ipsa."

As a hypothetical, take that Uber fatality but change the fact set around a bit to clarify the matter. In this hypothetical, the pedestrian was in a crosswalk, and had surviving family to sue. If this is a product liability case, then Uber's investigation into what happened is extremely relevant (giving Uber an incentive to hide or manipulate the conclusions) and this is a massively expensive case to prosecute. This is probably worth it in a wrongful death action, but wouldn't be for comparatively minor injuries. If this is a negligence case, then it is straightforward: a tragedy, but a well understood legal procedure.

The other question, to the legal mind, is who is liable? There is the old joke about the military that amateurs discussion strategy, while professionals discuss logistics. The equivalent in personal injury law is that amateurs discuss whose fault it was, while professionals discuss how much insurance is available. A wrongful death by a commercial vehicle, with a commercial insurance policy, is a different case from a wrongful death by someone with the state legal minimum liability insurance. (Yes, you can go after that person's assets. No, people with cheapo insurance don't have assets to go after.) But if it is Waymo or Uber that is liable, every case has deep pockets. This won't affect the minor injuries, but for the serious stuff is is huge.

In my article on accidents from a few years ago.

Discovering who was at fault will take just a few minutes. (It did not here because the police confiscated the car before Uber could dump the data logs.) Now everybody knows to program their car to upload as much data as possible over the cell networks if there is ever a crash where the safety driver is no longer fully in possession of the vehicle.

So it will all be about product liability, negligence etc. and it could be very expensive. And this was my point. If we get 1/10th the accidents but each one costs 50x, the system has screwed up, since it makes it create more liability to be safer.

Which is exactly what causes people to talk about arbitration and other elements in this law.

If discovering who was at fault will take just a few minutes, why would there be depositions of programmers?

The fault that will take a minute to determine is the usual police "fault in the accident" which is to say, who failed to yield right of way. In almost every accident there is somebody who failed to yield, and it is their fault. It's more complex than that but with the full 3D geometry of the accident on display, the legal fault will be very clear. Of course, the big deal is if the robocar was the one that failed to yield.

In that case, the next big question is, "why?" and in particular, what error in the programming or design caused it to fail. This is a different level of fault, and that's what you depose programmers over. The plaintiff's lawyer wants to show negligence, wants to show that the product is "defective" in some way. In extreme cases, that this was criminal or warranting punitive damages. (That's very rare.)

But it's expensive. Nobody asks to look inside the brain of a human driver to find out why they did not look left. They just didn't.

If the robocar failed to yield, presumably the company would admit negligence and a product defect, and there would be no reason for the plaintiff to depose the programmers to prove that. Maybe they could depose the programmers to try to prove punitive damages, but unless there's something egregious going on (like in the Uber case) I don't see it getting that far. Between the interrogatories and requests for production (especially emails) it should be clear enough that it's just a simple error (and not something like executives ordering programmers to turn off safety features) that doing a bunch of depositions would obviously be a waste of money.

They will do their own internal investigation, and if at fault, offer a generous settlement. That won't stop an eager plaintiff's lawyer from trying for more.

Depends how good the case is. But my point is that if they admit negligence, then there's no need for discovery to prove negligence. If the parties can't settle, the trial can focus solely on damages. This generally shouldn't require a deposition of programmers. It'll be pretty much identical to a non-robocar case. In fact, the defendant might even be able to keep the jury from even hearing that it's a robocar case, as this is irrelevant to the issue of damages in most situations.

When you say treat it as a negligence case, do you mean pretending that the robot is human and asking if the robot was negligent?

Because, otherwise, if the question is whether or not a human being was negligent, it's just as complicated as a products liability case, if not even more complicated.

If you do mean pretending that the robot is human and asking if the robot was negligent, what's the standard of care? A reasonable robot? :D

Any negligence or liability will lie with the company, for the actions of its employees (programmers and others.) Robots are not actors under the law, only humans (and companies being treated as fictitious legal persons.)

Sorry for the confusion. My question was to Richard, who suggested it was possible to treat a robocars crash like any other traffic crash case. But since, as you point out, we don't treat robots as humans, unless the law is changed we *can't* treat these cases like any other traffic crash case, at least not under a negligence theory. The negligence happens at a completely different time than in a normal traffic crash case.

DriveNow does this with their electric BMWs in Europe - if the car is low on charge (lower than some threshold like 20%) - finishing your trip with parking it on a charger gives you a 30minutes bonus. That way you can do a sort of free commute, but won’t receive any real monetary value, so no cheating with scooter-hoarding makes sense.

"It offers an clause forcing disputes to arbitration"

How do they accomplish that anyway? I'm not sure how the federal government can force arbitration in the absence of a contractual agreement between the plaintiff and the defendant, without violating the Seventh Amendment.

From what I'm reading, it looks like what previous bills did was *not prohibit* arbitration clauses in terms of service. That makes more sense (it wouldn't be binding on situations where a third party who didn't agree to the terms of service was the plaintiff). But maybe this newer bill actually forces arbitration rather than not prohibiting it?

Well, there are a few meanings of negligence here. There is the negligence of doing things like violating the vehicle code and failing to yield, which is the ordinary thing you see in a vehicle accident. Then there's the negligence of not properly testing software or having bugs, which is more on the product defect side.

It is an interesting question as to what would happen in the defendant said, "Yes, we agree our car failed to yield, and it did so because of the following bug in our software which was not caught because our testing procedures had no test to catch that, but we have added one." Would the plaintiff say they no longer seek to do discovery to look for other potential causes or errors? Yes, they need not argue too much about the damages because, as I have said, to avoid the costly case, I think settlement offers will be generous.

I'm not sure you understand "negligence" and "product defect" law. Only a person can be negligent. So the negligence in a robocar case would not be failing to yield, unless the robocar is considered a person (which I suppose is possible, as non-humans can be persons). But presumably the negligence would be negligently writing code in a way that doesn't yield (if you're going after the manufacturer). Or negligently operating a car knowing that it fails to yield (if you're going after the operator). Or something like that.

And product defect laws are separate from "negligence." In fact, the whole point of them is that you don't have to prove that anyone was negligent. You just have to prove that the product was defective. In that sense it's actually product defect law where all you might have to prove is that the robot failed to yield, therefore it was defective (in practice it might be more complicated, as some failures to yield might be considered not to be a defect; you could possibly need expert witnesses to testify about what failures to yield are acceptable vs. which ones are a defect).

As far as depositions, if the defendant says (as in, admits in a request for admissions or response to an interrogatory or something equivalently binding) "Yes, we agree our car failed to yield, and it did so because of the following bug in our software which was not caught because our testing procedures had no test to catch that," and if the bug is clearly an honest mistake and not something that would give rise to punitive damages, the plaintiff would almost surely no longer seek to do discovery on the issue of fault, as fault has been admitted and there's no need for a trial on that issue. In fact, the defendant would likely be able to get a protective order against such a deposition, as the issue of fault is irrelevant. The trial will be on damages, or an affirmative defense, or contributory/comparative negligence, or something else that isn't agreed upon.

Something you may not know is that a trial isn't necessarily held on all issues. If the parties agree that there was negligence, but disagree on damages, then a trial is held on the issue of damages. In such a case, the jury would in fact be *prevented* from learning the details about negligence except to the extent they are relevant to the issue of damages. This is quite common, actually, as damages are often more subjective than liability.

In all probability these cases will be treated like a case where a crash is caused by something like a brake failure. Did the operator know that the brakes were likely to fail (or did they fail to inspect the brakes)? That's how you'd go after the operator. Otherwise, you go after the car manufacturer, likely for a product defect (as it's easier to prove), though possibly for negligence if there are limitations on what damages can be recovered in a product defect case. A robocar might have an extra party to possibly sue, as the car manufacturer might be separate from the software manufacturer, but otherwise it's fairly similar.

Also they could try for strict liability (dangerous instrumentality) against the operator, and there are probably other state-specific and motor-vehicle-specific tort laws (including, in many states, no-fault laws).

Obviously a car itself can't be negligent. The operator can operate it in a negligent way. And when it comes to a product defect, the damages can vary based on the cause of the defect, and if the cause was negligent design or construction of the product.

Since the vendor will know the causes and circumstances, we're only in a trial if the plaintiff's lawyer is hungry for blood and convinced the plaintiffs to refuse the generous settlement offer of the defendant. Which it's hard to imagine did not come fairly early.

The operator can operate the car in a negligent way. If we're talking about a true self-driving car, that might mean pushing the "on button" in a negligent way (by pushing it at all, knowing or having reason to know it was defective). Or maybe the negligence was in maintaining the vehicle (not fixing sensors, for instance).

I'm not sure what you mean when you say damages can vary based on the cause. Unless you mean punitive damages. And again, negligent design or construction is different from a product defect action. Under a product liability theory, you don't have to prove negligence. If what you're saying is that damages available for negligence are different from damages available for product liability, that's true in some cases (due to tort reform). But the hypothetical is that the defendant is admitting to both negligence and a product defect.

There may be many reasons for a trial. Many of them are about things of which the plaintiff doesn't even have direct knowledge. A good example would be a claim for loss of consortium due to a back injury. Back injuries can be difficult to prove, and can have devastating effects on both the injured party and his/her family. There very well may be a trial on that, if the defendant(s) don't believe the plaintiff or don't agree on the value of the loss. Pain and suffering is another issue where there very well may be a dispute. Comparative negligence might also play a role (fault isn't always 100%; in fact in major robocar crashes it's likely that both parties will be partly at fault, and percentages will be argued over). And defendants can only go so far in being "generous." At some point they need to draw the line.

Another complication in the case of a robocars is that there will often be multiple defendants, since the operator and the software manufacturer may be different parties, and might both have deep pockets. That is sure to raise litigation costs, and to make settlement more difficult to boot, though the operator and software manufactureright have an indemnity agreement to make it a bit easier to handle (and settle). Other parties might also be sued. Hardware failures might be an issue, for instance.

In the case of Waymo, Cruise and Zoox it's likely to be single defendant, and so for many others. The idea of independent operators is some way out, and when it happens, there will be strong contracts defining where liability lies and providing indemnities.

Now, when privately owned self-driving cars come out, it will be different but most companies who have said anything have said they will handle all liability in a case where the machine was driving. This won't stop the owner from being named as a defendant but they will be a smaller party and indemnified and will have waived rights to control the trial, I suspect.

In addition, while pain and suffering etc. will play a role, I expect the instructions to the defendant's legal team will be "Offer a settlement that is sufficient so that it is very likely that the plaintiff's lawyer has no choice but to recommend accepting."

At least, in any case they don't want to go to court on. This will just be the cheaper option. In some cases they may decide, "Hey, we like the facts on this one so let's go to court and pay the cost."

Word is that Uber offered the family of Herzberg a fairly lowball amount, but they took it. Uber has a whole team that deal with accident settlements (usually of course involving human driver incidents) and apparently it's a well oiled machine. I would not have personally used that team for that event, but it worked.

You're right that it may be a single defendant for many cases. Tesla may be the most prominent exception at first, though they likely won't have true self-driving for a while. I don't expect that situation to last very long, though. It doesn't make much sense for the same company to focus on building self-driving car technologies and to focus on the logistics of running a self-driving taxi company. And invariably there will be lots of parts built by other companies - parts that will invariably fail and cause crashes. Whether or not these types of failures will be a significant portion of crashes is hard to say. One argument that they might be is that these crashes are most likely to happen when several failures combine together.

In some states (a minority of them), an insurance company may be a defendant, as well.

As far as contracts, they can provide indemnification, but if a third party who is not a party to the contract is injured, the contracts are useless in terms of defining liability and who can be sued. Along those lines, presumably arbitration clauses will be used for the people actually using the service, and in at least some states they should hold up (and possibly across the country if the federal government steps in to make them enforceable). But again, only parties to the contract will be held to them.

Pain and suffering etc. will play a huge role, as it already does. A defendant who tells its lawyers to always offer a settlement that is too high will find that the word quickly gets out that they are doing this and the necessary settlement will get higher and higher. On the other hand, offering a lowball amount at first is a common technique. It reveals little to no information ("whoever offers a number first, loses" should probably be changed to "whoever offers a reasonable number first, loses") and at least some of the time the lowball offer is accepted. I'd certainly expect defendants to continue doing that, but the question is not what their initial offer is, but how high they're willing to go if the plaintiff doesn't accept the initial offer.

In any case, I don't expect there to be a lot of cases that go to trial. Trials are very costly on both (all) sides, and each (every) side is almost always required to pay their own legal fees, win or lose. I do expect the values of settlements to go up significantly. Right now the vast majority of drivers are underinsured and undercapitalized to handle a serious personal injury lawsuit. Many cases get settled for the policy maximums. But as you point out, at least at first the operators of most of the self-driving vehicles are going to be big companies with deep pockets, so there will be no need to settle for the policy maximum. On the other hand, many states have enacted some form of tort reform that may have its own maximums in many cases.

I disagree. I think they should make a very public position that they make a generous offer, but they never increase it unless new information arises. If the lawsuit will be expensive, it's expensive for both sides (probably more for defendant) and everybody knows this, and the defendant's desire to keep it from court. It's worth the premium to avoid messy, public cases when you already know the facts and whose side they are on.

There's constantly new information. And that's one of the problems with laying all your cards on the table from the start. If the information you don't have is in your favor, the other side can accept the offer and you've overpaid.

The geometry of the accident will be worked out within minutes. The source of any error by the software will probably be clear in days. There should not be a lot new found on the cause of the particular accident. Some things like deeper bugs and bad procedures will take longer to figure out, but if you are a plaintiff, do you want to bet on the expensive task of finding them or the generous settlement?

The cause of the crash is not the only important information. What kind of "generous settlement" are you going to offer without knowing the extent of the injuries?

Why would you not know that? You would, of course, be talking in great detail with the plaintiffs to find out about that sort of thing.

This is starting to sound a lot like the usual process of trying to negotiate a settlement. I think you're oversimplifying it, though. Yes, the two sides will be in contact, but damages are often hard to prove. So will the company take everyone at their word when they describe the sort of damages they have? Some damages (especially pain and sufferinh) are very subjective. If you offer an amount that is highly likely to be accepted, you'll be vastly overpaying most of the time. In any case, we're not talking about something that will be known in minutes or a small number of days. Not in a lot of potential situations.

So I really don't see much difference between this and other cases. One difference, which will complicate things, is that there *will* often be multiple parties, as a two car collision will likely have at least one injured passenger in each vehicle. And that could become quite complicated in quite foreseeable scenarios: If a human driven car makes a mistake *and* a self-driving car also makes a mistake, injuring a passenger in the self-driving vehicle, now you have a complicated three party (or more) lawsuit, especially if the human driver has deep pockets just like the robocar company.

The goal will be to overpay. The plaintiff will know that.

As you know, the Uber fatality was settled within a week, so our one example seems to indicate counter evidence to the slow model.

If the goal is to overpay, I guess the only lawsuit you'll really have to worry about is the shareholder lawsuit. :D

The goal is not to overpay so much as to make the lawsuit go away because of the expense of a public battle. That's always a factor, but it's stronger here.

There's no benefit to making your first offer be your best offer that helps you accomplish that goal. In fact, doing so is counterproductive.

If you start out with a lowball offer, it usually won't be accepted. But if it is accepted, it saves you money in two ways: You save money by paying less, but you also save money because it's much easier to make a lowball offer; you don't have to do much fact-checking to make a lowball offer, whereas to make your absolute best offer you have to know all the facts about the case (not just cause, but damages, and probably more importantly, how effective the plaintiff will likely be at proving cause and damages).

Sorry for not threading. It's accidental. (Except for this time, as this is a different topic.)

Add new comment