Tesla Autopilot A Threat To Cyclists?

Tesla Autopilot


Tesla Autopilot

Tesla Autopilot

A Stanford University robotics researcher shares her two cents about Tesla’s features, and she seems less than thrilled about Tesla Autopilot.

Heather Knight was so unimpressed with the Autopilot’s ability to see bikes, that she wrote a paper on Medium entitled “Tesla Autopilot Review: Bikers Will Die.” Tesla Autopilot is not the only Tesla feature that was upsetting to Knight, but there were also a few features she did enjoy.

Update:  A follow-up/companion piece on Fortune had its title edited/softened to “Tesla’s Autopilot Tech Is a Danger to Cyclists, Robotics Expert Says”, as the media outfit stated, “The headline on this post was edited at 3:17 on May 29 to more accurately reflect the nature of Tesla’s semi-autonomous driving technology.”

Tesla Autopilot

Autopilot is not full self-driving software.

Knight holds a doctorate degree in robotics from Carnegie Mellon, and is currently studying social robotics. Checking on the Tesla Autopilot system was on her to do list. Another Stanford researcher accompanied her on a Tesla Model X test drive, to see just how well the car’s features live up to expectations. More specifically, how do they handle the human-robot interaction experience? She discovered the Tesla’s:

“Agnostic behavior around bicyclists to be frightening … put biker lives at risk.”

“I’d estimate that Autopilot classified ~30% of other cars, and 1% of bicyclists. Not being able to classify objects doesn’t mean the tesla doesn’t see that something is there, but given the lives at stake, we recommend that people NEVER USE TESLA AUTOPILOT AROUND BICYCLISTS!”

Thus far, there have not been any reported incidents between Tesla vehicles and bikes, but there was an accident in Norway involving a motorcycle.

The interesting part about all of this is that she says that the Autopilot works well in regards to cars. Yet, she says that it only identifies about ~30 percent of them? As she explains, the car attempts to show everything that the camera sees, but often it “sees” objects, but doesn’t always classify/display them on the screen. So, while it may make the driver/robot interaction a bit scary, the car is potentially seeing the bikes, but not showing the bikes.

People took to Twitter to comment to Knight about her findings. She made it aware that she is a fan of Tesla, but she wanted to point out some issues. She calls the technology a “human in the loop” system, which is exactly what Autopilot is said to be. Tesla has not yet released its full self-driving software, and it has made this abundantly clear. In Knight’s paper, she explains:

“The Tesla Autopilot feature is basically a button to turn the car into autonomous driving mode.”

This statement is true because the car functions at Level 2 autonomy. But, this could also be seen as a stretch if we are reading the word autonomous as “self-driving” (Level 4 autonomy).

Knight hopes that people will understand the distinction. Despite her criticism, and the morbid title, she gave an ‘A+’ to Tesla’s Situation Awareness Display, Automatic Lane Switching, and the Tesla App. She graded the Recessed Door Handles and Steering Around Curves with a ‘B.’ User Set Target Velocity and Giant Touchscreen get a ‘C’ from Ms. Knight, and she gave the Self-Locking Feature an ‘F.’ This is because they got locked out of the vehicle, but the Tesla App saved them.

To see her full detailed review, click the Medium link below.

Source: Fortune via Medium

Categories: Tesla

Tags: , , ,

Leave a Reply

49 Comments on "Tesla Autopilot A Threat To Cyclists?"

newest oldest most voted

So when I cycle around I should be looking over my shoulder to see if a Tesla is hunting me?

In my experiences, the biggest threat to cyclists are the cyclists themselves.

Alright! Let the lowest common denominator discussion ensue!


Godwin’s Law. Bam. 😉

Seriously though, I’m not sure I’m sooo worried about Autopilot around cyclists. From what it sounds like the system sees them perfectly fine, it just can’t reliably tell what they are. As a biker myself, as long as it drives around me (and gives me a reasonably wide berth), I’m happy to share the road with a computer. Plenty of human drivers are jerks too.

As for the risks to cyclists, studies generally show that the health gains from cycling more than offset the risks due to pollution and accidents.



Also, the more bikes and electric cars and the fewer infernal combustion engines there are on the road, the better off everyone will be. I think most people would agree with that. Maybe even Hitler 😉 And wouldn’t that confuse Mr. Godwin.

I absolutely agree.

In the same sense that the biggest threat to (car) drivers are the drivers themselves… but I guess that’s what Tesla is addressing! Self-driving bicycles, anyone?? 🙂

The Tesla driver who decapitated himself fits more in that category.

Have some respect for the dead.

RIP Joshua Brown !!!

You mean like the respect that the Florida Marlins showed to deceased pitcher Jose Fernandez who killed himself and his two friends while piloting his speedboat drunk and high on cocaine, by honoring him with a huge bronze statute. I wonder how the families of the two friends that he killed feel about all the respect shown to Jose Fernandez by the Marlins organization.

It’s more important to use that Tesla t-bone crash as a teachable moment, than to worry about hurt feelings. Sorry if that sounds harsh, but people need to be shown that certain actions/behaviors can have deadly consequences.


Yeah I’m just saying have some respect don’t be a jerk that is even possible

When you run one over, you can tell the Officer, “but he got in the way of my auto-pilot!”, and then the Officer can ticket the cyclist. 😉

“In my experiences, the biggest threat to cyclists are the cyclists themselves.”

Only if by “threat”, you mean a bicyclist’s willingness to share the road with large heavy fast-moving vehicles driven by inattentive people with a tendency to overlook anything smaller than a car.

If that’s not what you mean, then I think it’s safe to conclude you haven’t spent much time riding a bicycle on public roads.

Then your ‘experiences’ are flawed. Tell me, how many cyclists are killed or injured by other cyclists, as opposed to the numbers hurt by vehicles??

Try riding a bike and you will think differently. Also, the stats don’t back you up at all.

I rode a bike on roads when I went to University 45 years ago. One day a driver mowed down a group of cyclists on the wide and well marked shoulder of a local highway. The driver claimed that a cigarette dropped in his lap i.e. distracted driving. I never rode my bike on roads after that.

It’s gotten sad. Chris Frume, a TDF winner, Nicky Hayden, a successful Moto GP rider. Both were struck on bicycles, in public, in just the last couple months. It’s becoming so common, you have to be the best rider in the world to make the news.

People aren’t just more distracted. They are more belligerent. The Kalamazoo, MI, incident was just about one year ago. Anyone even remember a guy who took out 9 cyclists, killing 5, in one hit?

RE: Tesla. It’s as bad with cyclists as it is with the other offset impacts one can find on Youtube. Fortunately, Tesla’s AP is bad enough in two-lane, off highway situations that just about no one uses it.

Wow, I had no idea Nicky Hayden died just a week ago from bicycle crash. Guy who raced 2 wheeler at 200 MPH died from going probably less than 20 MPH. Sad to hear another legend is gone.

As someone who values his life I never ride my bike on a public road. it’s just not worth it. Last weekend a cyclist was killed by a driver where I live and a couple days ago a motorcycle rider died just down the street from my house.

I’ll take my air conditioned cage and be alive any day over being an organ donor in waiting…

You could also become an organ donor in that cage.

You take a calculated risk for any activity, and same is true with driving, bicycling, motorcycling, going to the beach. Probably the safest place would be a gun range, but you can’t live there.

The better answer is to engineer roads to separate bikes and cars, autonomous cars or otherwise. Separated bike tracks, like separate sidewalks for pedestrians have been around for generations. It’s not rocket science.


True a calculated risk. While it is best to stay home, when your options are car vs bike and the car is 10 times safer, the calculation favors the car. That is the calculated risk – calling it calculated just reinforces the point.

I see many, many Ghost Bikes in NYC. These memorials are white-painted bikes that mark the location where a bicyclist was killed or seriously injured by a car.



Solution is simple: autopilot only on highway where bikes are not allowed. There’s way too many “stuff” in local roads for autopilot, for now anyway.

Being a cyclist and motorcyclist myself, I do worry about AP’s ability. It’s not only cyclists, but small kids and animals as well.


As a cyclist and an AP user myself, I’m not worried about this “problem” because it is in fact a human-in-the-loop system and it really doesn’t lend itself to being used on cycling streets anyway. While there’s no doubt it could be misused this way, the kind of people who will do so are also the kind of people who don’t pay proper mind to cyclists when driving by hand, so I don’t really see it making matters worse. Better if anything, because at least it sees bicycles *sometimes*.

I would be willing to put a special reflector on my bicycle that would glow brightly to Tesla’s on-board cameras. Tesla could give them away as promotional items – in the shape of Tesla ‘T’ .

It’s occurred to me several times that it wouldn’t be a bad idea to make bikes more visible to radar — google “corner reflector” or “radar reflector”.

Problem is, as far as I’ve been able to determine, although a radar reflector is cheap, it’s also somewhat large compared to a bicycle, order of 30 cm on a side. I wonder if a more compact solution could be devised, but I think the physics of the situation mean it would have to actively transmit (so, expensive) unless the car’s radar wavelength is quite short (anyone know what wavelength cruise control radars such as Tesla’s uses?).

If you aim a IR device (like a TV remote control) onto a camera – it makes a large reflection on camera image – even if it is invisible to human eyes.

So maybe installing IR lights on back of a bike – and any other object – will make the object stand out.

All of this is a roundabout way of trying to cope with the design decision of not installing LIDAR on your car.


What we need isn’t to make the bicycle more reflective. What we need is for the autonomous car to have good enough sensor and software systems to unfailingly spot (not “identify”) the position and movement of all moving obstacles large enough to contain even a small baby; obstacles including bicycles and their riders.

And anyway, the radar/lidar reflector would only work within a relatively narrow angle. We need the autonomous car to unfailingly spot the bicycle no matter which direction it’s headed.

Once autonomous cars are properly engineered, bicycle riders will be far safer sharing the roads with them than with human drivers.

Thanks for explaining, but as a bicyclist I’ll be the judge of what I need. And what I would rather have is safety equipment that works with the fleet currently on the road (which includes plenty of cars that aren’t Teslas which are filled with front facing radars, by the way). Wishing that cars had LIDAR, plus $2.50, will get me a cup of coffee but won’t make me any safer in the foreseeable future.

Certainly it would be wise to increase your bicycle’s reflectivity to radar. However, if your comment about reflector size is correct, then any flat object that large would significantly increase the wind resistance of your bicycle, which would be a substantial disadvantage, and hence would make the marginal added safety of rather questionable benefit.

I apologize if I came across as being unconcerned for your safety. I’ve ridden a bicycle on public roads myself, in the past, enough to understand the very real concerns of those posting here about the issue of bicycling safety.

It’s not at all a trivial issue, but in respect to autonomous driving I see it as more of a subset of the general issue of avoiding collisions with people not inside automobiles, whether they’re pedestrians, or riding bicycles, scooters, or motorcycles. Autonomous vehicle designers, and self-driving software engineers, can’t focus just on bicycle riders as a stand-alone category. There needs to be a more general solution to detecting and avoiding collisions with people on the roads but not inside cars or trucks.

That will remain true whether or not you, personally, find a practical way to increase your bicycle’s reflectivity to radar and/or lidar.

I’m all for improving things going forward, and yes, a radar corner that looks like a 30 cm aluminum foil Christmas tree ornament won’t be practical. (Though as noted the required size is dependent on the radar wavelength, and for that matter perhaps other geometries might also be usable, e.g. a metallized mylar backed backpack?)

It would be interesting to know things like, what percentage of cars on the road have forward-looking collision avoidance radars already? If it’s just a small fraction of a percent, it’s probably not worth going to great lengths to defend against them, but if it’s into double digits, maybe there’s a market opportunity for someone to devise an active beacon, similar to the v2v transponder you and @carcus were discussing downthread, but not requiring v2v. (But again, I don’t know the practicality of this — possibly vehicle radars would view this as a spoofing attack and already incorporate countermeasures.)

As an aside, speaking of v2v I’m somewhat skeptical that we’re really going to see v2v deployment in 2021, unless the industry starts taking security seriously.

Toyota has taken the first steps in implementing a vehicle-to-vehicle communications system over a dedicated frequency. In the future, this technology might trickle down to bicycles, and even the smartphones of pedestrians crossing a street that would communicate with vehicle when the smartphone’s GPS determines that a person is crossing a street for example.

Toyota became the first automaker to “mass market talking cars—that is to say, vehicles that can receive and share data transmitted by external infrastructure and by other vehicles.”

“So-called ‘vehicle-to-infrastructure’ and ‘vehicle-to-vehicle’ communications are key to enabling autonomous cars because they provide essential safety information that sensors and cameras onboard vehicles cannot pick up.”

Toyota offers an “Intelligent Transportation System” (ITS) safety package on three of its vehicles sold in Japan. “The ITS technology uses a frequency of 760 megahertz, which is standardized for Japan, to send and receive information between vehicles and roadway infrastructure.”


(⌐■_■) Trollnonymous


I’d prefer my “Model à trois” come without it for $2,500.00 less.

Sadly that won’t happen.

Wow that was a terrible article, but at least the text of the article is much more responsible than the clickbait headline.

One point the article does indirectly make is that it can be hard to get used to a new car. But this is generally true of many cars as anyone who has rented one in the last ten years knows, now that cars have having UI that diverges from key/steering wheel/pedals/shift lever. Yep, it’s a pain to get in the driver’s seat of any random vehicle and have to spend time figuring out where to put the key, where the shifter is, and so on.

Maybe the author thinks this is unique to Tesla (or TESLA or tesla as she insists on calling it, I guess she has a problem with operating her shift key too) because as an impecunious graduate student she has never rented a car and had to deal with unfamiliar controls? Or has never rented a car with controls more complicated than a 1970 Dodge Dart?

If the Tesla really did classify cars 30% of the time as she estimated, there’d be tons of Tesla Autopilot on Youtube.

Questionable or not a scientific study from a professor.

She seems to be suffering, to a greater or lesser degree, from a case of ivory tower myopia; locked into her academic tunnel vision viewpoint of a tidy world in which chaos and entropy don’t exist, one where perfection is attainable and theory always equals practice.

In theory, there’s no difference between theory and practice. But in practice, there is.

“Questionable or not a scientific study from a professor.” O rly?

I think you mean, “questionable or not, a blog post from a graduate student.”

Two observations:

1) Pedestrians and bicyclists will die whether or not AP exists, and have been dying long before Tesla was ever even a company. The original title was grotesquely provocative.

2) Her analysis has been debunked elsewhere around the net. There is no evidence at all that more cyclists have been hit with AP turned on. Much less that they are more at hazard with AP than without it. Like most faulty “analysis” of AP, it simply ignores the statistical realities. The only actual statistical analysis of AP showed that it greatly improved safety.

No safety system will eliminate all accidents 100% of the time. If AP reduces accidents with bicyclists by even just 30%, that’s actually a win.

You never hear anything about it —

, but it seems likely that all road going vehicles (including bicycles) will eventually be required (by law or insurance) to have an operable transponder (as an important part, but certainly not the only part of vehicle recognition) before autonomous vehicles come out of their “experimental” stage.

“Coupled with autonomous technology, it may have further lifesaving effects. Although Google didn’t use V2V communications for a landmark driverless journey around Austin, Texas, in a car that didn’t contain steering wheels or brake pedals, many experts believe that these communications will enhance, and perhaps even be necessary, for operations at the most advanced levels of autonomous driving.”

Feds Want V2V Communication in New Cars Starting in 2021

carcus said:

“…it seems likely that all road going vehicles (including bicycles) will eventually be required (by law or insurance) to have an operable transponder… before autonomous vehicles come out of their ‘experimental’ stage.”

Unlike all too many suggestions regarding autonomous driving, this one actually makes sense, would significantly improve safety, and could be implemented in the real world.

Yes, I think this might actually happen.

I’m not buying into this concept, at all. A fully autonomous vehicle should be able to locate roads, lanes on those roads, and obstacles (or potential obstacles) in its immediate environment, and observe which of those obstacles is moving, and in what direction and speed. Aside from other vehicles on the road, there is no need to “identify” (in the sense of “figure out its identity”) any one of those obstacles; the robotic system needs only to avoid colliding with any of them, and as far as possible, maintain a safe distance from moving objects. How sad that someone who holds a doctorate degree in robotics doesn’t get that. She appears to be buying into the fallacy that self-driving cars actually need to interpret and understand reality on a human level. No, they need to observe and interpret reality on a much simpler level, so they don’t get confused and distracted by complexities not necessary for safe driving. I don’t want my hypothetical self-driving car to be wasting time and computing cycles trying to tell the difference between a bicycle rider and an empty cardboard box being blown along by the wind. Optical object recognition systems simply don’t work reliably… Read more »

Here’s a few statistical questions not answered.
1. How many road miles have autonomous Teslas done.
2. How many cyclits have they hit.
3. Is that more or less than average.

When it comes to whether its dangerous or not how the system actually works is less relevant then actual real world results.

You can call a system as dangerous as you like but if real world results prove otherwise then it’s just not as dangerous as you think, the reverse can be equally true.

As a lifelong cyclist, I used to look forward to when we would run out of oil. Unfortunately, we are too clever and keep finding ways to wring out a bit more. Now I look forward to the collapse of the financial shell game that allows drilling for “unconventional” oil at a loss.


HR interactions specialist is NOT expert on test drives.

Wonder why that person penned those articles. Wont improve her standing. There is… scientific method missing.

As a cyclist I don’t fear autonomous driving cars more than I do fear human drivers. In fact autonomous driving cars probably won’t insult me for making use of my right to use the same lane as the cars do.