Next generation autopilot - Model S vs XC90

[Ohji]

I am excited to take delivery of my Model S sometime in the next couple of months, as Autopilot 2.0 was far more than I was expecting from Tesla at this time. That being said, I thought it would be interesting/fun to discuss the differences between Volvo's approach and Tesla's, seeing as both chose the same autonomous driving computing platform.

While both systems are powered by the Drive PX2, the sensor suite chosen by each is quite different:

Tesla: 8 cameras, 1 radar, 12 ultrasonic sensors
Volvo: 5 cameras (including 1 "trifocal" camera), 7 radar (4 surround radar and 3 long range), 12 ultrasonic sensors, and 1 LiDAR

Also, while we aren't yet sure what the redundancy is in the Tesla, Volvo states that all critical systems have a backup including the computer, sensors, steering, and brakes.

It seems that Volvo has a more robust sensor suite whereas Tesla is going to be more camera-dependent.

The details of the Volvo solution are described here:
Self-driving car test | Intellisafe | Volvo Cars

Drive Me, the world’s most ambitious and advanced public autonomous driving experiment, starts today

My worry is that Tesla may not have the redudancy necessary for regulatory approval of their L5 platform, and they will be forced to issue an updated autopilot hardware suite to conform to regulations. Here's an example: if one of Volvo's side cameras fails or gets blocked with snow/debris, the radar should be able to serve as a reasonable backup to keep the car safe to pull over or let the driver take control. The Tesla, on the other hand, has no side radar and so would need to rely on the very limited ultrasonic sensors in the same scenario.

From what Elon has said, Tesla has tested this platform for a year, so I am sure there is more to the story. For Tesla to advertise full autonomy, there must be everything necessary for level 5 in the car. Hopefully we will learn more about how the system works and the backup systems in place as time goes by.

 

[calisnow]

This, like all other comparisons thus far is not valid because Volvo is vaporware and Tesla is shippable product. A limited run 100 car test in specially designated zones of Copenhagen or London does not count as shippable product in my book.

Currently back in the real world Tesla has the most accurate and reliable autopilot system in existence - and they are doing it with one camera and one radar. The only remotely competing system is Mercedes with multiple cameras, multiple radars - and the MBZ still can't even find the lines.

Now, Tesla has made their own, best-in-the-real-world autopilot obsolete with 2.0. Their lead is only accelerating. As Tesla has shown it is the software, not the sensors, that make the system.

When Volvo builds something you can buy, what they are doing will be relevant and we can make comparisons then.

 

[calisnow]

...Volvo states that all critical systems have a backup including the computer, sensors, steering, and brakes. ... My worry is that Tesla may not have the redudancy necessary for regulatory approval of their L5 platform, and they will be forced to issue an updated autopilot hardware suite to conform to regulations. Here's an example...

Yes - time will march on and Tesla will build better cars with better sensors and computers. As for your example it boils down to "Non-existent-vaporware-x would be superior to actual-product-Y in the following scenario." If we play that game I can go further. My vaporware will have 100 cameras, 10 PX2's, and rocket jets to make the car leap 100 feet in the air should something unexpected happen up ahead.

 

[Cosmacelf]

I'm not sure that Volvo sensor hardware suite is the same one they would use in a production car. I suspect they over configured those 100 vehicles, which are R&D testbeds, so see what they actually need in a production car. They are still in the R&D phase...

 

[calisnow]

I'm not sure that Volvo sensor hardware suite is the same one they would use in a production car. I suspect they over configured those 100 vehicles, which are R&D testbeds, so see what they actually need in a production car. They are still in the R&D phase...

I would go further - Volvo is in the PR phase. Notice that Tesla doesn't go around pulling marketing stunts like making a highly publicized deal with cities claiming to be doing R&D. They just quietly do their R&D and then ship a product you can buy. To me, this kind of thing with Volvo smells of "Hey give us a chance folks - look, look we're working on it! Pay no attention to that product Tesla is selling, or their exploding sales. Their shippable product is an unsupervised wannabe! Our vaporware is the real thing - just look at all these sensors it has. And remember we are the safety brand, not them. Don't forget that! Hey - where you going?"

 

[HumanGenome]

You are forgetting that the Volvo is a showroom PR vehicle partnered/paid marketing stunt by Uber and not a production vehicle.

That hardware suite is not making it to Volvo production consumer vehicles..

 

[green1]

Hardware wise many manufacturers are way ahead of Tesla, and Tesla's hardware is embarrassingly lacking. That said, the Tesla software is pushing the envelope father than anyone else.
The end result is an interesting race. Tesla's AP2.0 will never make a fully autonomous vehicle. It's simply not capable of it. But maybe, just maybe, it will be able to do what they promised autopilot would do 2 years ago.
The Volvo on the other hand has the hardware to do full autonomy with ease, but no useful software to back it up, and likely won't catch up to Tesla's current autopilot offering for several years yet.

Combine the 2 and you'd be getting somewhere. As it is, I think we're still a very long way from owning our own self driving cars.

 

[Muhammad]

Hardware wise many manufacturers are way ahead of Tesla, and Tesla's hardware is embarrassingly lacking.

Just because Tesla doesn't throw in 50 ugly sensors on every corner of the car doesn't mean Tesla is not capable of doing so. In fact, the work Tesla is doing with camera recognition and radar is well ahead of everyone else

I believe Tesla's engineers who are the best in the world at autonomy over what anyone else on these forums say about what hardware is necessary for a safe self driving car. Try putting all that Volvo hardware on a production car and you'll quickly realize it will be entirely prohibitive from a financial standpoint. Also, to program all those sensors to truly and accurately work together will take multiples greater of time and the chances of software glitches and corner cases increases.

I think Tesla is redundant enough to be safe "in most instances." If the front camera fails, you still have the front radar which will formulate a point cloud akin to lidar of the road in front and the front/side camera should be totally operational by virtue of their ingenious position. I clean my car by hand and can attest to the fact that all dirt and water is guided along the bottom sides of the car and that portion between the windows in particular doesn't get very dirty at all.

Overall, I agree with Tesla's approach to this. Use the least amount of hardware possible and focus your efforts on the software. The fact that humans can drive solely through optical means is a testament to the very little hardware actually required for autonomy

 

[calisnow]

...Tesla's AP2.0 will never make a fully autonomous vehicle. It's simply not capable of it...

What is it lacking, in terms of sensors and computing power?

 

[GOFORIT5]

I love how there are so many on the forums that seem to know the caliber of Tesla engineers or the extent to which AP2 will be capable.

I hope Tesla and Volvo and anyone else comes out with more and more advanced autonomous vehicle developments because to me it seems like it will make mine and a lot of others' lives better.

The speculation and the know-it-all comments are still fun to read so don't let this post discourage you from adding your two cents

 

[SwTslaGrl]

I think AP2 hw could be sufficient for several years. Perhaps in the future combined with some vehicles broadcasting their GPS position, like emergency vehicles, heavy trucks, roadworks, stopped cars etc.

 

[malcolm]

Next generation?

I wasn't aware that Volvo had succeeded with first generation pedestrian detection, let alone some form of AP.


Volvo blames horrible ‘self-parking car accident’ on driver (from May 2015)

And I can't resist this from 2010

 

[J1mbo]

IMO Tesla is lacking rear radar for full autonomy with redundancy. Visual and ultrasonic capability are both impacted by dirty sensors, rain & snow; visual is also impacted by direct sun, low contrast & fog.

An example would be the car pulling out of a parallel parking space into a busy street - would the car always see the cyclist or moped that that is approaching, without fail, regardless of weather/time of day/etc? A human driver relying on vision and audio* can fail in this situation, but the expectation is being set that autopilot is orders of magnitude safer than a human driver.

* The cyclist can ring a bell or shout to attract a human's attention. No such luck with any of the autonomous solutions that I am aware of!

 

[gangzoom]

I've been driving for 15 years+, to my knowledge haven't killed anyone and only been in one accident. I have no radar, ultrasound sensors, and a very limited field of view with my foward facing eyes. If you compare the hardware I have access to whilst driving to AP 2.0 there should only be one winner... But I do have access to the best image processing software on the planet .

 

[green1]

I've been driving for 15 years+, to my knowledge haven't killed anyone and only been in one accident. I have no radar, ultrasound sensors, and a very limited field of view with my foward facing eyes. If you compare the hardware I have access to whilst driving to AP 2.0 there should only be one winner... But I do have access to the best image processing software on the planet .

You also have windows that are far larger than your eyes. Try being limited to complete loss of vision if even 1cm of your windshield has dirt on it. That's the issue. The cameras are NOT as good as your eyes because their field of view is drastically limited.

I agree with others that LIDAR is ridiculous, it has all the downsides of both cameras and radar, with none of the upsides of either. But you have to realize that cameras are not a simple replacement for your own eyes. They have many limitations that your eyes do not.

Cameras are necessary forward facing to recognize lines on the road, signs, people directing traffic, etc. a camera is necessary rear facing to detect overtaking emergency vehicles. But for all other purposes radar is far better than camera. Tesla even admits as much when they talk about how capable their forward radar is at seeing through fog, rain, etc. It's funny that they don't realize that same necessity for all other angles of the car, ESPECIALLY being that only the front camera has any way of clearing the glass in front of it.

AP2.0 MIGHT get to level 3 autonomous driving, but without the ability to drive in inclement weather it will never make level 4

 

[TexasEV]

To the OP, you're neglecting the most important comparison. Tesla Model S and X are EVs. Volvo XC90 has an ICE. Enough said.

 

[Max*]

What is it lacking, in terms of sensors and computing power?

I think he's alluding to redundancy, because it doesn't seem that the Tesla has much of it in place. From the picture of the placement of the cameras, if one of the cameras goes out, there are portions that are not overlapping which may not be enough to satisfy the full autonomous driving.

 

[Gizmotoy]

Next generation?
I wasn't aware that Volvo had succeeded with first generation pedestrian detection, let alone some form of AP.
Volvo blames horrible ‘self-parking car accident’ on driver (from May 2015)
And I can't resist this from 2010

They're funny, but those are both ridiculous examples, right?

In the first, the car wasn't equipped with the feature being "demonstrated." That's hardly the fault of the vehicle. That's like me ramming my Classic P85 into a wall and saying "Tesla's Automatic Emergency Braking doesn't work!!"

For the 2010 video, the system is disabled below 2mph. That assumes the car was even on. Given the dinging when the door is open the car appears to be off and rolling in neutral with the key in the ACC position. Safety systems are disabled when the engine is off. It looks like the first failures were probably from sensor blindness with the pedestrian starting so close to the vehicle, though. A person coming around the side of the car to walk in front of it while it's moving is not exactly a use case you're likely to encounter. It's also probably worth noting they've shipped 3 generations of the system since then and the upgraded system coming on the XC90 makes 4.

Full disclosure: We have a V60, which in the front fuses data from a camera, three radar units (one of which is long range), and a LIDAR unit. I can confirm it gets very angry if you try to go when there are pedestrians nearby. It'll detect a pedestrian approaching a crosswalk and chirp, and if they stop so too does the chime, but if they step out into your path the HUD on the windshield goes nuts. Pretty cool, actually. In balance, I can also confirm its highway lane holding is laughably bad. Like, it should not even be an option it's so bad. Ping pong city.

 

[bradhs]

It's all in the software and processing power. What is Volvo using for processing? Plus, Tesla knows how to highly advance software.

Is Volvo relying on third parties like MobileEye?

 

[Bladerskb]

In terms of redundancy Tesla's no hardware just falls flat. I think it will have a difficult time getting through regulations