Be mindful when interpreting self-driving truck tryouts.
I like a feel-good story as much as the next person.
There has been a recent spate of media gushing about self-driving semi-trailer trucks, heralding the vaunted arrival of driverless big trucks due to one or another automaker or tech firm having made a publicity related stunt involving a highway journey debut, including usually carrying along a noteworthy load of produce or products from point A to point B.
This kind of feat oftentimes gets pundits into a lather of speculating that self-driving capabilities for large-scale trucks are nearly ready for prime time and implying that the self-driving tech involved is akin to a Level 5 fully Autonomous Vehicle (more on this in a moment).
The most recent news-making trek occurred last week.
Reportedly, a semi-trailer truck hauling about 40,000 pounds of butter went on a three-day journey across the United States, starting in Tulare, California and ending in Quakertown, Pennsylvania, and did so (apparently) predominantly in driverless mode as equipped by Plus.ai, an up-and-coming self-driving vehicle startup company based in Silicon Valley.
The delivery was a real one in that Land O’Lakes had wanted the butter hauled from California to Pennsylvania and so the journey accomplished two birds in one stone, namely delivering the needed butter and showcasing driverless capabilities for trucking and giving Plus.ai some media attention.
According to news accounts, there was both a safety back-up driver and a system engineer in the truck throughout the trip.
The safety back-up driver would have likely been in a position to rapidly take over the truck if a disengagement was needed (note that “disengagement” is industry parlance for when a human back-up driver takes over the wheel from the AI system that’s driving). The ride-along system engineer likely monitored the internal status of the on-board computers and AI, regularly scanning the sensory input being collected and monitoring the AI’s status during the trip.
Allegedly, the journey took place without any disengagements.
The reason that the trek took three days, which is longer than the time it would take to straight-out have the AI drive non-stop across the U.S. (no need for the AI to rest or get coffee), apparently was to give the human back-up driver the federally mandated breaks required when doing truck hauling.
Reporters touted this butter delivering journey as an incredible feat that demonstrates the advent of true self-driving trucks.
Many commentators also held this accomplishment over the heads of the self-driving car makers, trying to make the driverless car builders appear to be behind-the-times and dis them as dismal laggards in comparison to driverless big haul makers.
As I said, I like a feel-good story, but it would also be helpful if the media didn’t rush to judgment, including having the press render outlandish comments or unsupported declarations that are fanciful and not in keeping with the facts.
Let’s unpack the matter.
The Levels Of Self-Driving Vehicles
It is important to clarify what I mean when referring to true self-driving vehicles.
True self-driving vehicles are ones that the AI drives the car or truck entirely on its own and there isn’t any human assistance during the driving task.
These driverless vehicles are considered a Level 4 and Level 5, while a car or truck that requires a human driver to co-share the driving effort is usually considered at a Level 2 or Level 3. The vehicles that co-share the driving task are described as being semi-autonomous, and typically contain a variety of automated add-on’s that are referred to as ADAS (Advanced Driver-Assistance Systems).
There is not yet a true self-driving vehicle at Level 5, which we don’t yet even know if this will be possible to achieve, and nor how long it will take to get there.
Meanwhile, the Level 4 efforts are gradually trying to get some traction by undergoing very narrow and selective public roadway trials, though there is controversy over whether this testing should be allowed per se (we are all life-or-death guinea pigs in an experiment taking place on our highways and byways, some point out).
Since semi-autonomous vehicles require a human driver, the adoption of those types of cars or trucks won’t be markedly different than driving conventional vehicles, so I’m not going to include them in this discussion about self-driving big hauls.
For semi-autonomous vehicles, it is equally important that I mention a disturbing aspect that’s been arising, namely that in spite of those human drivers that keep posting videos of themselves falling asleep at the wheel of a Level 2 or Level 3 car, we all need to avoid being misled into believing that the driver can take away their attention from the driving task while driving a semi-autonomous car or truck.
You are the responsible party for the driving actions of the vehicle, regardless of how much automation might be tossed into a Level 2 or Level 3.
Self-Driving Vehicles And The Truth
For Level 4 and Level 5 true self-driving vehicles, there won’t be a human driver involved in the driving task.
All occupants will be passengers.
A self-driving vehicle at a Level 4 is less capable than being at a Level 5.
Generally, Level 4 is a self-imposed scope delineation in terms of where and when the self-driving capability will function, referred to as the Operational Design Domain (ODD) for the vehicle.
For example, I might make a self-driving AI system that can only cope with sunny weather conditions and works exclusively within a specific city boundary. That would be the ODD for any vehicle using my self-driving tech. If it is raining, the AI would refuse to drive the vehicle, and likewise, if you wanted to go outside the city bounds the AI would indicate it was unable to do so.
Level 5 is essentially a no-ODD constrained form of self-driving tech. A true Level 5 would be able to drive anywhere that a human could drive a vehicle, including in any kind of weather or place that a human was able to do the driving.
In the case of the butter delivery story, the Plus.ai company has indicated that they are currently testing at a Level 4, not a Level 5 capability.
I mention this facet since the mass media doesn’t oftentimes understand and nor convey the difference between being at a Level 4 versus at a Level 5 (it’s a huge difference!).
Now, don’t misunderstand my point as though I am somehow knocking the efforts of achieving a Level 4.
I’m decidedly not knocking it.
Just about everyone in the self-driving realm is aiming right now at Level 4, including the self-driving car makers and the self-driving truck makers, doing so as the cornerstone step toward ultimately reaching the vaunted Level 5.
Accomplishing Level 4 is a tremendous step in of itself, so don’t falsely denigrate it as somehow just a minor stepping stone.
Getting to true Level 4 for any kind of expansive ODD is a well-sought accomplishment and will have substantive benefits for providing self-driving capabilities to society.
My carping about this butter story is that the media implied or at times stated that true self-driving has arrived, which is false and misleading on several accounts:
· First, this was only a tryout run and it is mistaken to extrapolate the experiment into saying that self-driving has been solved.
· Second, this and other such tryouts are at Level 4, and not a showing of Level 5, thus the experiment is about ODD-constrained self-driving and not about the elusive full self-driving.
Why are such nuances important?
The media is fanning confusion among the public and regulators, doing so by declaring prematurely that full and complete self-driving has been achieved, and the next day trouncing self-driving and offering the opinion that it won’t ever occur.
These kinds of good cop and bad cop spasms of storytelling about self-driving aren’t helping anyone other than the press that wants to get heightened ratings for their news reports (for my tips on how to assess videos posted about self-driving vehicles, see this link here).
I’d also like to tackle the comparisons being made or implied about cars versus trucks in the self-driving realm.
Does a seemingly self-driving truck that goes on a cross-country trek mean that somehow the big haul self-driving is light-years ahead of the self-driving car tech?
No.
There is nothing magically better or supreme for self-driving truck tech than there is for self-driving car tech.
It’s pretty much the same on an overarching basis.
A key difference between self-driving for trucks versus cars is more so about the nature of trucks and what they do, rather than the self-driving tech itself.
I’ll explore that next.
Self-Driving Trucks And What They Do
Semi-trailer hauling trucks are big and heavy vehicles, while cars are generally much lighter and smaller in size.
I think we can all agree to that presumption.
Highway driving is usually markedly different than say downtown street driving, notably due to the fact that highway driving involves primarily straight-ahead driving with an occasional lane change, and thus there aren’t as many turns or complex driving as would be found in a downtown area.
Furthermore, highways are usually devoid of pedestrians, bicyclists, and others that might get in the way of driving.
I’m not saying there is never such obstructions on highways and just pointing out that street-level driving typically involves an ongoing, persistent, and intensified cacophony of disruptions and potential object collisions that need to be continually coped with.
As such, when you test self-driving tech on a vehicle that’s on the highway, you aren’t necessarily simultaneously testing as though that tech can also handle driving on side streets and downtown areas.
In some respects, a car or truck that you put onto a highway can almost be like a train, staying within its lane for long periods of time, and making various lane changes only as needed. Trains have the luxury of being on a railroad track and so don’t need to do much in terms of being able to drive, at least not in the sense of human driving per se.
Whenever you hear about a self-driving vehicle that was tested on a highway, you must ask whether that tryout included only highway driving or also includes off-highway driving too.
Recall the infamous Uber Otto self-driving semi-hauler truck that made a 120-mile run to deliver beer (see my debunking of what happened).
For some of these self-driving truck tryouts, the human back-up driver is utilized to get the truck to the highway, and once the highway driving is begun the AI is switched on. Then, after going whatever distance or time might be involved, the human back-up driver turns-off the AI and drives the truck off the highway to a rest stop or to the targeted delivery warehouse.
Therefore, when headlines blare that a self-driving truck delivered milk or pajamas, we don’t really know whether the self-driving was employed only while on the highway or also on the side streets leading to and from the highway.
Plus, even if the side streets were used, another handy trick is to have pre-mapped just the stretch between the starting point and the highway entrance, so that the AI is prepared for that first part of the trek. And, you can aim to do that portion in the early morning hours, before much traffic has developed on the side streets.
All in all, as might be evident, there are plenty of ways to make the trek much easier for the AI and yet the media fails to ask questions and probe as to the details of the journey that was performed.
A banner headline that says a big haul truck delivered a load of peanuts is raft with assumptions such as the AI of the truck was able to deal with the difficulties of getting onto the highways and off the highways, when needed, as needed, along with having driven on the highways themselves.
We don’t know that to be the case.
Another facet of big truck driving is that usually a human truck driver tries to minimize the number of lane changes and stays in the slow lane of traffic (often as mandated by regulations).
Again, this is unlike car driving in that a human car driver often gets into and out of lanes on a frequent basis. They opt to dart over to the fast lane, then change lanes to get around a car here or there and are navigating in and around the rest of the traffic.
There is a lot of discretion in driving.
I remember that my grandparents would get onto the freeway and stay entirely in the slow lane. They never ventured out of it. They clung to the right side of the freeway with dear life. During their time on the freeways, they were counting the miles and minutes to reach the off-ramp they would be taking.
I think we can all agree that driving in such a limited fashion is different in tone and complexity than when freewheeling among the beehive of freeway traffic.
Yet another facet about semi-trailer trucks is that they are big and imposing.
Much of the time, cars keep away from the big trucks.
Wherever the truck wants to go, cars get out of the way. It’s like a big dog that causes the little puppies to scatter.
This can make things easier for the AI.
The AI might be driving a big haul truck with not much finesse and relying upon the bulkiness and fierceness of the imposing truck to spur the rest of traffic to get out of its way.
I realize that any experienced truck driver will have angst about this characterization of highway driving and point out that car drivers do lots of stupid and dangerous things, such as cutting in front of a big haul or riding on the bumper of a big haul.
Yes, those are absolutely valid aspects that involve truck driving. And, dealing with the complicated physics of a lumbering truck that can’t turn on a dime is definitely tricky in its own right for the AI. Trucks are faced with longer stopping distances and less agility than cars.
But we don’t know that any particular tryout of a self-driving big truck on a highway involved those kinds of occurrences and testing.
Maybe so, maybe not.
In fact, believe it or not, there have been self-driving truck highway stunts whereby those undertaking the effort had brought human-driven cars onto the highway that road next to the truck, serving as guardians while the truck made its trek. Those defensive position cars would attempt to keep other traffic from getting in the way of the truck.
Smart?
Sneaky?
You decide.
Nobody readily knew about these hidden angel efforts and so the media went along with assuming that the truck has entirely done the trek on its own, yet the reality was that a plethora of guide vehicles driven by humans surrounded the truck during its journey in a kind of safety cocoon.
Conclusion
There are several ways that the media could do a better job of reporting on self-driving truck efforts.
One aspect would be to find out the underlying details of a stunt or trek that is being reported on.
Besides an indication of how many disengagements there were, let’s not stop with just that singular number since it can be somewhat meaningless depending upon how the driving task was structured and undertaken.
Questions abound such as:
· Did the trek involve the AI driving the entire journey, end-to-end, including when off the highway too?
· What was the complexity of the traffic and driving while off the highway?
· While on the highway, how much of the time did the truck stay in its lane and do straight-ahead driving?
· How many lane changes occurred?
· What was the complexity of the lane changes?
· How many times did nearby car traffic attempt an incursion into the truck driving and in what ways did the AI deal with those incursions (did it overly rely on its bullying factor to prevail over the cars)?
· What was the number of encounters with other trucks and navigating around say slower moving trucks or otherwise contending with fellow trucks nearby (presumably human-driven trucks)?
· Did the truck encounter highway debris or other unexpected and unplanned difficulties, and if so, how did the AI respond?
· Did the AI system abide overtly by the state-by-state driving regulations as it went cross-country?
· Were there any close calls and if so, how did the AI driving system react?
· Was the truck alone or did it have added help from allied support vehicles during the trek?
· Etc.
Some suggest that these trucking tryout efforts ought to have an unbiased third-party “auditor” that goes along during the trip, being able to witness first-hand what happens, doing so to augment whatever claims are later made about the journey.
By the way, the third-party shouldn’t be someone that can have the wool pooled over their eyes, which oftentimes is done by asking a junior reporter that doesn’t know squat about self-driving tech to go along.
Finally, there are critics that argue these big truck tryouts on our highways are the worst kind of guinea pig experiments of self-driving tech.
Allow me to elaborate.
Having a human back-up driver during a self-driving vehicle tryout is not a guarantee that a crash is not going to occur. Take a look at the sad and now famous incident of the Uber self-driving car in Arizona that rammed and killed a pedestrian at nighttime that was jaywalking, and you’ll see that the human back-up driver was apparently not paying attention to the driving task at the time of the incident.
It’s one thing to have a car go awry and cause an accident while having a large semi-trailer truck do so could be even worse.
The human back-up driver in the truck might be caught unawares after hours upon hours of not driving and be lulled into believing that the AI is doing just fine. They might be delayed in responding to a rapidly emerging dilemma.
Some question whether the self-driving truck (and self-driving car) efforts are prematurely using our pubic roadways, and ought to be focusing instead on more testing at proving grounds or via simulations, waiting until better prepared to use our open highways and byways.
Anyway, the odds are that we are going to continue to have highway tryouts of both self-driving trucks and self-driving cars.
The self-driving truck stories are eye-catching since we all tend to think of driving a truck as a rough and tough chore and if a truck can do self-driving its awesome to imagine, perhaps more so than driving an everyday and mundane car.
Of course, human truck drivers aren’t as enamored of the looming day of AI driving trucks, but it’s a day that will eventually arrive, presumably reducing the costs of hauling goods and making our roads safer, assuming that the AI is appropriately readied and capable to do the driving.
Self-driving tech firms and automakers that carryout public tryouts of big haul trucks can help society understand the current state-of-the-art, limitations included, doing so by being upfront in a detailed manner about how they are undertaking their tryouts and how much more needs to be done.
The ads that urge people to go to truck driving school and earn a living via driving trucks will eventually give way as AI takes the wheel.
We decidedly aren’t there yet.
