Tesla's autopilot, on the road

My Nissan Leaf Forum

Help Support My Nissan Leaf Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
C&D not impressed with AP's progress:

Where Are Autonomous Cars Right Now? Four Systems Tested

We go hands-off in four vehicles you can buy today to understand the current state of automated technology and why progress won't come easily.


...A year after its debut, Tesla’s Enhanced Autopilot still offers no perceivable enhancement over the first-generation Autopilot. Compared with the original, the second-generation hardware suite employs four times as many cameras, ultrasonic sensors with double the range, and 40 times the computing power, yet Enhanced Autopilot shipped in October 2016 without fundamental abilities such as self-steering and adaptive cruise control because Tesla was still developing the software. All the sensors in the world can’t guide a car if the computer can’t make sense of the world, and Tesla took a temporary backward step when it brought Autopilot development in-house after an ugly public breakup with Mobileye, the supplier behind the original Autopilot.

By this past August, when we shepherded a $146,200 Model S P100D around Southern California, Enhanced Autopilot operated at the same level as the first-gen system, both according to Tesla and our own observations. The self-steering algorithm is no more certain or reliable than the earlier version we lived with in our long-term 2015 Tesla Model S. The technology that oozed possibility two years ago now simply feels incomplete, especially since Enhanced Autopilot makes the same unsettling mistakes as its predecessor did—mistakes that suggest a Tesla can’t always make sense of its environment.

Autopilot can cover hundreds of miles flawlessly or it can make a dozen errors in as many miles. In either case, we’re always caught off guard when the system makes an abrupt steering input that attempts to yank the Model S out of its lane without warning...
https://www.caranddriver.com/features/where-are-autonomous-cars-right-now-four-systems-tested-feature

For example:

https://www.youtube.com/watch?v=XxAoS4VvpjQ&feature=youtu.be
 
I guess it's reasonable to find the current AP2 un-impessive compared to C&D's previous test using AP1.
I take it as a compliment to Tesla that C&D feels it's as good as AP1. That was Tesla's big challenge when they split with MobileEye (now Intel).
That Tesla's AP tech is judged to have equal performance with tech from a company valued at $15B is rather impressive still.

I've driven both AP1 and AP2 and have many hours with AP2. Yeah, it needs better lane keeping and awareness. I think that's a fair criticism.
I watch it like a hawk.
The other EAP feature, TACC works great IMO. Better than the BMW TACC system I've driven. I'm never hesitant to employ TACC on even short trips and it has accurately warned me of impending collisions (simultaneous with my own "uh oh!").

I will order EAP with my Model 3. It's that good.
 
sparky said:
I guess it's reasonable to find the current AP2 un-impessive compared to C&D's previous test using AP1.

Right, AP2 software is still based on the original AP1 developed with MobilEye and now with software enhancements
provided by the transition to the Nvidia processor. Most likely Tesla just used the MobilEye AP1 source code
and re-compiled it with minor updates to run on the Nvidia processor. Lack of processor capabilities, e.g. MIPS,
was one the limitations for AP2 and as such, staying with MobilEye.

sparky said:
That was Tesla's big challenge when they split with MobileEye (now Intel).
That Tesla's AP tech is judged to have equal performance with tech from a company valued at $15B is rather impressive still.

So what, they've been now relying on Nvidia, i.e. almost a year now, that gave AP2 the updated processor
capabilities for the additionally needed AI processing since MobileEye left. That's hardly a "big challenge"
nor a singular Tesla achievement!
 
Autopilot software problems aside, I've been wondering why TSLA's hardware inadequacies have not been brought up.

Until now:

‘Full of crap’ -- Tesla’s Elon Musk just got seriously dissed by a GM exec

Published: Oct 6, 2017 2:16 p.m. ET

...Miller — who helps direct GM’s efforts in autonomous driving — isn’t buying a key Tesla claim...

Musk & Co. have said all Tesla vehicles produced since October 2016, including the $35,000 Model 3 sedan, contain “the hardware needed for full self-driving capability,” though such features aren’t currently enabled. That implies the vehicles are what the industry calls “Level 5” autonomous cars.

“I think he’s full of crap,” said Miller, according to multiple reports. “To be what an SAE Level 5 full autonomous system is, I don’t think he has the content to do that.”...
http://www.marketwatch.com/story/full-of-crap----teslas-elon-musk-just-got-seriously-dissed-by-a-gm-exec-2017-10-06

On topic thread:

Elon Musk is 'full of crap'
https://teslamotorsclub.com/tmc/threads/elon-musk-is-full-of-crap.99534/
 
edatoakrun said:
Autopilot can cover hundreds of miles flawlessly or it can make a dozen errors in as many miles. In either case, we’re always caught off guard when the system makes an abrupt steering input that attempts to yank the Model S out of its lane without warning...
https://www.caranddriver.com/features/where-are-autonomous-cars-right-now-four-systems-tested-feature
Thanks for posting this article. It is very interesting to see four L1 / L2 systems compared. I wish ProPilot had been included.
 
jhm614 said:
edatoakrun said:
Autopilot can cover hundreds of miles flawlessly or it can make a dozen errors in as many miles. In either case, we’re always caught off guard when the system makes an abrupt steering input that attempts to yank the Model S out of its lane without warning...
https://www.caranddriver.com/features/where-are-autonomous-cars-right-now-four-systems-tested-feature
Thanks for posting this article. It is very interesting to see four L1 / L2 systems compared. I wish ProPilot had been included.
Yes, an excellent article, and another confirmation for me that I'll wait for Level 4 before trusting my life to such a system. As long as I've got to keep an eye out for occasional surprises that the system doesn't know how to deal with, I need to remain fully engaged with the driving in order to have the necessary reaction time.

As it is, all of the systems allow too much time to pass without driver engagement before bringing the car to a stop - Joshua Brown had the semi in sight for only about 10.4 seconds after cresting the hill, which was plenty of time to avoid it or stop if he'd been paying attention, but none of these systems would have required him to start paying attention until well after impact, even assuming they could detect the semi crossing the road. In the interim, AEB provides an extra layer of safety to a fully engaged driver.
 
lorenfb said:
sparky said:
I guess it's reasonable to find the current AP2 un-impessive compared to C&D's previous test using AP1.

Right, AP2 software is still based on the original AP1 developed with MobilEye and now with software enhancements
provided by the transition to the Nvidia processor. Most likely Tesla just used the MobilEye AP1 source code
and re-compiled it with minor updates to run on the Nvidia processor.
As I understood it, AP2 did not start with Mobileye software, so I don't think any of that is true.
 
jlv said:
lorenfb said:
sparky said:
I guess it's reasonable to find the current AP2 un-impessive compared to C&D's previous test using AP1.

Right, AP2 software is still based on the original AP1 developed with MobilEye and now with software enhancements
provided by the transition to the Nvidia processor. Most likely Tesla just used the MobilEye AP1 source code
and re-compiled it with minor updates to run on the Nvidia processor.
As I understood it, AP2 did not start with Mobileye software, so I don't think any of that is true.

So you think that Tesla scrapped the total AP1 development effort and all the software logic/code and
started from scratch when they switched to the Nvidia processor? Do you realize the lost time and
AP1 engineering costs would basically have had to be written-off? It's not as if AP1 was a simple Windows
API or DLL. Yes, major mods were required for the transition to AP2, but not a total ground-up totally
new development.

Hey, maybe I'm wrong (Elon is known for different management approaches). Please provide your source.
 
I'm not up on the details (and probably couldn't understand them anyway), but here's one article I found googling Tesla Mobileye software, which AIUI indicates that Tesla is going its own way:

https://electrek.co/2017/03/13/tesla-vision-autopilot-chip-intel-mobileye/

Here's another article: https://cleantechnica.com/2017/05/0...months-recreate-mobileye-chip-tech-elontalks/

Of course, what Elon says and what actually happened may or may not be the same thing. How much they could carry over remains to be seen.
 
A good summary of the business and legal implications of the “Enhanced Autopilot” fiasco:

Elon Musk Was Wrong About Self-Driving Teslas

He’s charging $8,000 for a feature that doesn’t exist yet while owners bemoan a current version that’s flawed. This is the story of how we got here.


A year ago, Tesla Inc. set up a hastily organized conference call between Elon Musk and reporters. Tesla had something big to announce—big enough for the chief executive to open the floor to questions. That doesn’t happen very often.

What followed wasn’t so much the unveiling of a new product as a plan for a product. Tesla’s driver-assistance platform, Autopilot, was about to begin a transformation to fully autonomous driving. Every Tesla would come with eight cameras, radar, 12 ultrasonic sensors, and a Nvidia Corp. supercomputer. Once testing and regulatory approval were complete, Musk said, the car would be able to drive entirely by itself.

“The foundation is laid,” Musk proclaimed. Tesla was so confident, in fact, that it started selling its “Full Self Driving” feature for an additional $8,000 on any new Model X or Model S. Tesla’s timeline was, as is so often the case, years ahead of what most believed possible...

What followed were months of setbacks, delays, and in-house turmoil. A year later, there’s still no sign of Full Self Driving, and even the less ambitious hasn’t quite reached parity with an earlier, discontinued version. The head of Tesla’s Autopilot division left in January, and six months later his successor did, too. Meanwhile, Tesla owners who paid thousands of dollars for the options filed a class action lawsuit, alleging they were tricked into buying a feature that doesn’t exist and—in some cases—an unsafe car...
https://www.bloomberg.com/news/articles/2017-10-25/can-tesla-make-up-for-autopilot-s-lost-year
 
I think the tenor of the Bloomberg article is a simple notion that FSD was a bridge too far. Tesla hasn't delivered.
If I paid $3k for FSD last year, I think Tesla owes me a refund or something of equal value.
I did pay $5k for EAP and I'm enjoying it. Not wishing for a refund at all. Will select the EAP option on my Model 3.
 
Via IEVS:
Tesla Autopilot 2.5 Activates High-Speed Automatic Emergency Braking
https://insideevs.com/tesla-autopilot-2-5-activates-high-speed-aeb/

The maximum speed at which Automatic Emergency Braking is available has increased from 50 mph to 90 mph.”

The newest Autopilot software update also adds back the vehicles’ folding side mirrors and reportedly better Autosteer. The mirrors will stay folded at 30 mph or less if you’ve opted to already fold them. The notes explain:

For ample clearance when driving through narrow streets, if you’ve folded your vehicle’s mirrors, they will stay folded while you’re driving at low speeds (up to 30 mph).
 
GRA said:
Via ABG:
Tesla on Autopilot plows into fire truck at 65 mph; NTSB investigating
https://www.autoblog.com/2018/01/23/tesla-autopilot-crash-fire-truck-ntsb/
The article below explains why Autopilot probably would not have prevented that crash, or any other crashes into stationary objects, even if it is operating perfectly as designed:

WHY TESLA'S AUTOPILOT CAN'T SEE A STOPPED FIRETRUCK

..Raj Rajkumar, who researches autonomous driving at Carnegie Mellon University, thinks those assumptions concern one of Tesla's key sensors. “The radars they use are apparently meant for detecting moving objects (as typically used in adaptive cruise control systems), and seem to be not very good in detecting stationary objects," he says...

The long term solution is to combine a several sensors, with different abilities, with more computing power. Key amongst them is lidar. These sensors use lasers to build a precise, detailed map of the world around the car, and can easily distinguish between a hub cap and a cop car. The problem is that compared to radar, lidar is a young technology. It's still very expensive, and isn't robust enough to survive a life of hitting potholes and getting pelted with rain and snow. Just about everybody working on a fully self-driving system—the kind that doesn't depend on lazy, inattentive humans for support—plans to use lidar, along with radar and cameras.

Except for Elon Musk. The Tesla CEO insists he can make his cars fully autonomous—no supervision necessary—with just radars and cameras. He hasn't proven his claim just yet, and no one knows if he ever will...
https://www.wired.com/story/tesla-autopilot-why-crash-radar/
 
The Model 3 also runs into parked firetrucks: Elon Promises Safety Upgrades After Model 3 Suffers Severe Crash Using Autopilot:
CleanTechnica said:
Later, Reddit user mikhpat added these details. “The driver of the Tesla is my dad’s friend. He said that he was behind a pickup truck with AP engaged. The pickup truck suddenly swerved into the right lane because of the firetruck parked ahead. Because the pickup truck was too high to see over, he didn’t have enough time to react. He hit the firetruck at 65mph and the steering column was pushed 2 feet inwards toward him. Luckily, he wasn’t hurt. He fully acknowledges that he should’ve been paying more attention and isn’t blaming Tesla.
Apparently, Tesla's autopilot system is designed to run into stopped firetrucks in this case:
CleanTechnica said:
Wired on Thursday asked this question: “How is it possible that one of the most advanced driving systems on the planet doesn’t see a freaking fire truck, dead ahead?” Wired dug into that topic and came up with this answer: Autonomous driving systems are programmed to ignore stopped vehicles. Volvo offers a system very similar to Tesla’s Autopilot. It’s driver’s manual warns people that the system can actually accelerate into a stopped vehicle if the car ahead suddenly swerves out the way. The traffic aware cruise control will ignore the stopped car and try to get back up to the designated speed set by the driver.

The reason is that if cars went into full “anchors away” panic braking mode every time their sensors detected a stationary object in the road ahead, they would be slowing most of the time and at risk of being rammed from behind themselves. Software engineers deliberately design these systems to ignore stationary objects in order to eliminate the problem of false positives.

Tesla covers this explicitly in its owner’s manual. “Traffic-Aware Cruise Control cannot detect all objects and may not brake/decelerate for stationary vehicles, especially in situations when you are driving over 50 mph (80 km/h) and a vehicle you are following moves out of your driving path and a stationary vehicle or object is in front of you instead.”
Sorry, but that doesn't make a bit of sense to me. The reason is that even if emergency braking results in being rear-ended, it does two important things:
1) It alerts those behind you of impending danger.
2) It might save the occupant of the Tesla or those in or near the stopped vehicle ahead.

This is the third case I know of where a Tesla has run into a stopped emergency vehicle. It strikes me that Tesla's autopilot is not obeying "Move Over" laws and is therefore endangering emergency personnel. Fortunately no one has been killed in these accidents.

Since Tesla has not addressed this issue, it seems that the NHTSA needs to step in to protect emergency workers from this danger.
 
edatoakrun said:
The drive to idiot-proof Autopilot, thwarted by fruit.

Time to turn on TSLA's driver-observation-camera...

https://www.youtube.com/watch?v=Hl9y7v8UohM
This jerk who doesn't care about others sharing the road w/him was looking for the perfect weight to bypass the nags: https://teslamotorsclub.com/tmc/threads/autopilot-steering-wheel-hack.108331/.

https://teslamotorsclub.com/tmc/threads/autopilot-steering-wheel-hack.108331/#post-2559580 was his response to my reply. I'm glad at least everyone (so far) "disagreed".
 
Tesla has released the next version of Autopilot 2.0 to users for beta testing:
Electrek said:
The bigger difference is the improvements to existing Autopilot features, like Autosteer, due to a much more advanced neural net system to power the Autopilot’s computer vision.
One question: Since the Tegra (X1?) processor doesn't not have a "deep-learning accelerator" like is found on the upcoming Xavier processor, are the neural networks in the Tesla Autopilot all implemented in the GPU?

This will be good news for Tesla owners:
Electrek said:
This new update is likely going to fix that. Two drivers we spoke to said it was now equal or better than Autopilot 1 for highway driving.
 
RegGuheert said:
[One question: Since the Tegra (X1?) processor doesn't not have a "deep-learning accelerator" like is found on the upcoming Xavier processor, are the neural networks in the Tesla Autopilot all implemented in the GPU?

The answer is yes given the extensive processor demands, i.e. the enormity of data, of the neural network system being implemented
by Tesla for AP without using LIDAR. Tesla most likely will switch to the later Nvidia processor when available as system demands approach
AP 5. The Nvidia stock was a great buy at about $80 -$90 (now $245) when Tesla switched from Mobileye.
 
Back
Top