UPDATE: Euro NCAP Releases Tesla Model S Autopilot Video: Stops For Object

OCT 18 2018 BY STEVEN LOVEDAY 22

So much for Tesla vehicles not “seeing” or stopping for stationary objects.

Hopefully, we’ll have much more information soon regarding European NCAP automated testing for the Tesla Model S and other Tesla vehicles. For now, NCAP has released a preliminary video. A link is supplied to look at further details, but sadly, it requires a login, which InsideEVs doesn’t have access to. Nonetheless, one of the most important controversies surrounding Tesla Autopilot is tested and revealed.

We don’t know for sure what has changed and why this test is so much different from those initiated in the past, but clearly, the Tesla Model S “sees” the stationary car and alerts the driver and initiates automatic emergency braking. This is a huge step for Tesla vehicles — and for all vehicles for that matter — since the current/previous technology in almost every current vehicle was not programmed to stop in such situations. Still, there are many variables involved and this is simply one test. Drivers should not trust Tesla Autopilot or any other driver assistance system. Remaining engaged and alert, as well as following all the automaker’s precautions is always a must.

Tesla Model S Achieves 5-Star Euro NCAP Safety Rating (Images + Graphics + Video)

When we are able to access the more detailed new data and/or have more information to share, we’ll update this article or provide a new article with additional details. For additional information on Euro NCAP automated driving tests, click here.

Video Description via Euro NCAP on YouTube:

Euro NCAP 2018 Automated Testing: Tesla Model S Autopilot

In addition to the video, the Euro NCAP website includes the following comments:

‘Autopilot’ on the Tesla Model S gives the driver a high level of support with the vehicle primarily in control in both braking and steering scenarios. This results in a risk of over-reliance as, in some situations, the system still needs the driver to instantly correct and override the system.

The name “Autopilot” implies a fully automated system where the driver is not required. However, the limited scenarios tested clearly indicate that is not the case, nor is such a system legally allowed. The handbook mentions that the system is intended only for use on Highways and limited access roads, but the system is not geofenced and can therefore be engaged on any road with distinct lane markings. The legally-required hands-off warning requires no more than a gentle touch of the steering wheel to avoid system deactivation, rather than ensuring the driver is still in control. To avoid misuse, Tesla has implemented a so-called ‘one-strike-you-are-out’ where Autopilot is not available for the remainder of a journey if the driver fails to nudge the steering wheel occasionally.

In the braking tests, the Model S shows full braking support by the system in nearly all scenarios except for the cut-in and cut-out scenarios where there is limited vehicle support. The full system support in the stationary scenario may result in over-reliance. However, in the cut-in and cut-out scenarios, the driver is required to apply the brakes in due time, which may reduce the driver’s over-reliance on the system.

In steering support, the Tesla does not allow the driver to input any steering himself and the system will provide all the steering required in the S-bend scenario. When system steering limits are reached, the vehicle will slow down to make the turn, again eliminating the need for driver input. In the absence of lane markings, Autopilot will stay engaged and will try to steer a safe path. However, with the sensors the Tesla has, this is nearly impossible to do reliably and implies to the driver that the vehicle can take all corners which, again, may result in over-reliance.

Overall, the Tesla system is primarily in control with a risk of driver becoming over-reliant on the system.

Take a look at how Tesla Autopilot compares to Nissan ProPilot based on the NCAP videos:

Let us know your thoughts in the comment section below.

Categories: Tesla, Videos

Tags: , , ,

Leave a Reply

22 Comments on "UPDATE: Euro NCAP Releases Tesla Model S Autopilot Video: Stops For Object"

newest oldest most voted
Manuel

What kind of additional info do you need?
Their website has a pdf and YouTube video for every vehicle tested.
https://www.euroncap.com/en/vehicle-safety/safety-campaigns/2018-automated-driving-tests/

Tesla comes out as the best system, so much that there is fear the driver might rely to much on Autopilot.

It would be nice if someone could make a side by side video comparison with the other competitors.

Magnus H

“Tesla comes out as the best system, so much that there is fear the driver might rely to much on Autopilot.”

No, the fear is due to the car acting like it is in control, which increasee the risk for over reliance. Quote “Overall, the Tesla system is primarily in control with a risk of driver becoming over-reliant on the system.”

viriato

This systems are Level 2 of autonomous driving, so they vill be considered as a background safety system that improve the driver habilities, never as a fully capable autonomous driving system. In the best case, the most advanced technology available today is level 3.
Only level 4 is admited as a real AD system on road, and level 5 in any situation.

I understand the concern of the safety authorities about this, and all of us have seen lot of videos of people with a irresponsable behavior in cars with only level 2, accidents and sadly, casualties.

deine Mutter

That should have been the last hurdle for the Model 3 to come to Europe. I always assumed that they wouldn’t send it over before they can beat the safety assist category especially since the mobileye cars like the Nissan Leaf score full points.

Matt

Tesla has always worked for tracking moving objects that come to a stop. The problem for any of the systems is when the approach a car that is just sitting in the road such as a fire truck!

Pushmi-Pullyu

Right. And apparently the problem occurs only when the car with ABS – Tesla car or not — is moving at higher speeds. (Above 35 MPH?) At lower speeds, ABS systems do a better job of “seeing” stationary objects. Altho confusingly, ABS also seems to be shut off at very low speeds, perhaps <15 MPH.

The variables seem to be:

1. Whether or not the other car was stopped when the car with ABS first "saw" it. If it was moving and then came to as stop, the ABS system should react to it.

2. Whether or not the car with ABS is above the speed at which it shuts off detection of stationary objects in the vehicle's path.

3. Whether or not the car with ABS is moving above parking-lot speeds, perhaps 15 MPH or more. Below that threshold, ABS is shut down.

4. Whether the object in the vehicle's path reflects a proper radar signal. Very low-density materials, such as styrofoam and other types of plastic foam, probably won't register. Fuzzy, fluffy objects, such as toy stuffed animals, may also not register.

(continued…)

Scott Franco

Again #1 violates the physical law of relativity. A moving car cannot tell if objects are moving, or if it is moving. The radar does not *care* if the object is moving. If you are postulating that the car calculates that the object is stopped by subtracting its own speed from the radar derived calculation, then you are saying Tesla built the software to be irrational, which I doubt. #2 is the same thing over again there is no such animal as “detection of stationary objects”. Go back to physics 101.

#4 is correct, and that is the central issue here.

Henry

“A moving car cannot tell if objects are moving, or if it is moving. ” – that is incorrect. What you think relates to 2 objects in space without any point of reference.

The car is a smart object, it knows where it is and whether it is moving or not (GPS, speedometer, …). With sensors (radar, camera) it can also tell whether the other car is moving or not based on its own location, velocity and trajectory.

The key part is how well the sensors and the car’s computer work together to discern an object in its path and takes appropriate actions.

Pushmi-Pullyu

(…continued)

So the conclusion presented in this article, which seems to be “Thank goodness Tesla has fixed this”, is pretty much completely wrong from start to finish. This problem has several variables, as I’ve outlined above, and unfortunately far too many articles try to oversimplify the problem.

The problem hasn’t been fixed, and it’s not just Tesla that has this exact problem. It’s just Tesla that has a problem with auto accidents being reported as “news” when they aren’t reported as news for any other brand of car.

Henry

My Model 3 always warns me if I approached a parked car on the side of the road around a bend. It sees the parked car as an object directly in my path.

In traffic, it also sends warning when a car in front suddenly slows down (not to a full stop), but does not warn when creeping along a very low speed (5-10 mph or slower) in stop and go traffic.

I never want to take any chance with the emergency automatic braking, but it appears the warning systems works as intended.

bro1999

Did they do any tests with red fire trucks?

Pushmi-Pullyu

It’s the same problem with cars made by General Motors. The real problem here is that people expect Tesla cars to perform better, because Tesla Autopilot clearly has superior functionality as compared to any other semi-autonomous system in any other car… such as Cadillac Super Cruise.

Sadly, Tesla’s ABS system doesn’t work any better than any car made by other auto makers… including General Motors.

https://arstechnica.com/cars/2018/06/why-emergency-braking-systems-sometimes-hit-parked-cars-and-lane-dividers/

Cypress

😂 you really like to poke at the bear don’t you.

tyler

if a bag blows onto the road, will the car stop? should the car stop? how big of a box will make it stop?

Kenneth Bokor
Looks like good improvement by Tesla! However, just to be clear in this “comparison” (since the Leaf is shown as well): Telsa (all models): The new set of sensors going into Tesla vehicles include eight surround cameras for 360-degree visibility with as much as 250 meters of range, 12 updated ultrasonic sensors to add redundancies to the cameras, and forward-facing radar (which Tesla recently enhanced to serve as a primary control sensor alongside Tesla’s vision system). Nissan (Leaf-ProPilot): ProPilot utilizes four cameras, 12 sensors, and a front wave radar to control speed, following distance, and stopping. For ACC and AEB functions, only the one forward facing vision-based camera and the one forward radar are in play. IMO, Pro-Pilot does a pretty good job with a lot less tech and compute power that Tesla offers. Tesla is still one of the best systems for these types of avoidance situations, but you pay for it with their higher costs. In my MY2018 Leaf, the PP system has worked surprisingly well for me and actually has exceeded my expectations, even though I can echo the results of some of the NCAP results when the driver has to take over to avoid stationary object… Read more »
Scott Franco

Still missing the essential information: is this object transparent to radar. The ultrasonic detectors are only good for a meter or so. The radar is good for 10 meters or more (based on my personal observation on the highway). That means there is not enough time at highway speeds to stop based on the ultrasonic detectors, they are strictly low speed detectors.

It is easy to make an inflatable car reflect radar, adding aluminum foil will do it. The previous vids that showed the Tesla failing to brake for an inflatable car were pure BS, but of course without information, so is this one.

I postulate nothing has *changed*. What we need is actual scientific tests carried out by people who aren’t idiots and back up their data with facts. The rest is just sensationalism.

Terrible Tim

They publish detailed information, seems quite scientific to me

file:///Users/danaweick/Downloads/euro-ncap-aeb-c2c-test-protocol-v201.pdf

stimpacker

Not here to argue who has the best system.
Personally, I don’t trust any system based on the videos and known self-driving accidents to-date.

The cut-out scenario is the most dangerous.
Don’t get why the computer is blindly following the vehicle ahead and when that vehicle serves out of a stationary object (e.g. concrete divider or red fire truck), the computer won’t stop and just plows right into it.

My wish for self-driving technology is not to take corners on twisty roads (it’s more fun when I take the corners) or navigate through San Francisco (like Cruise, not running a cab service). I want to be able to get on the freeway for a long distance trip, set Autopilot, then nap a bit. The car will wake me up when it approaches my destination exit. Same for whenever my freeway trip has gridlocked traffic.

We’re almost there with Tesla’s Autopilot today but it’s not good enough. With lives at stake, good enough is not the enemy of perfection. Perfection should be the only goal.

Cypress

Would argue that it needs to be significantly better than humans, but not perfect.

Jason
If you pay attention to the AutoPilot display you can clearly see the Tesla identifies the cars every time. What seems to take a long time in the cut out test the Tesla takes a long time to identify the car is not moving. So my take away from this is the sensor frequency is not high enough or the processing power is not fast enough. If the sensor frequency was high enough then within a few ms the system would calculate the relative velocities and know to slow down. Same with the processor speed, if it can crunch the details fast enough then within a few ms it would know the relative velocities and take action. To me it looks like the system is running about 10 frames per second, so in this situation between one frame and the next you already lost 100ms, then if it takes 3 frames to work out relative velocities you already lost 300ms. It’s almost as bad as a human driver. If it is calculating 60, or even 30, frames per second then you would think within 10ms it has determined there is a problem and started to avoid it. Also, interesting that… Read more »
Cypress

“Overall, the Tesla system is primarily in control with a risk of driver becoming over-reliant on the system.”

Yep. This is my general concern with all these systems.