Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Blog Musk Touts ‘Quantum Leap” in Full Self-Driving Performance

This site may earn commission on affiliate links.


A “quantum leap” improvement is coming to Tesla’s Autopilot software in six to 10 weeks, Chief Executive Elon Musk said a tweet.

Musk called the new software a “fundamental architectural rewrite, not an incremental tweak.”






Musk said his personal car is running a “bleeding edge alpha build” of the software, which he also mentioned during Tesla’s Q2 earnings. 

“So it’s almost getting to the point where I can go from my house to work with no interventions, despite going through construction and widely varying situations,” Musk said on the earnings call. “So this is why I am very confident about full self-driving functionality being complete by the end of this year, is because I’m literally driving it.”

Tesla’s Full Self-Driving software has been slow to roll out against the company’s promises. Musk previously said a Tesla would drive from Los Angeles to New York using the Full Self Driving feature by the end of 2019. The company didn’t meet that goal. So, it will be interesting to see the state of Autopilot at the end of 2020.

 
Last edited by a moderator:
Is the advent of standardized V2V communication (5G application?) a catalyst to getting to L5, or will we just do the harder challenge first (human and machine driving intermixed), then eventually see more machine-to-machine problem solving?

Just my personal opinion. I think we need to "solve" the harder problems first involving human and machine driving intermixed. For one, we will have mixed human and machine driving for a very long time. Also, autonomous cars need to be able to plan and anticipate the behavior of other vehicles/pedestrians on the road whether we use V2V or not. But eventually, when 5G becomes ubiquitous, I think we will see a lot more V2V communication. As I see it, V2V is not required for L5 but it will make L5 a lot easier and safer. So it will be used at some point.
 
Just my personal opinion. I think we need to "solve" the harder problems first involving human and machine driving intermixed. For one, we will have mixed human and machine driving for a very long time. Also, autonomous cars need to be able to plan and anticipate the behavior of other vehicles/pedestrians on the road whether we use V2V or not. But eventually, when 5G becomes ubiquitous, I think we will see a lot more V2V communication. As I see it, V2V is not required for L5 but it will make L5 a lot easier and safer. So it will be used at some point.
Agree.

I guess a better question is, will V2V evolve to a level that every car communicates with all other cars, and if so, before or after L5? Such standardization and requisite V2V should happen imo, for safe, but also for other applications. But I could see it being more achievable in a place like China. Obviously rogue or malfunctioning cars “off the grid” would need to be handled by L4/L5.
 
  • Helpful
Reactions: diplomat33
Agree.

I guess a better question is, will V2V evolve to a level that every car communicates with all other cars, and if so, before or after L5? Such standardization and requisite V2V should happen imo, for safe, but also for other applications. But I could see it being more achievable in a place like China. Obviously rogue or malfunctioning cars “off the grid” would need to be handled by L4/L5.

Well, the average age of a car on US roads is about 12 years. So even when new cars get V2V, there will be older cars without V2V for awhile. It is also possible that all cars will be required to be retrofitted with V2V at some point. But that will take time and some older cars may not be compatible for the upgrade. Achieving V2V on all cars will take years.

I suspect we might see pockets of V2V in major cities first before it spreads to other areas. I think we will see L4 robotaxis in cities with V2V first.
 
will V2V evolve to a level that every car communicates with all other cars, and if so, before or after L5? Such standardization and requisite V2V should happen imo, for safe, but also for other applications.
The big thing that needs to be solved is perception. Pair that with good driving policy and you have a winner.

Every car already communicates - by being where it is and going as it is going, and if you eliminate human drivers you have visual indicators, a la turn signals, brake lights, etc.
With perception, you will have excellent accuracy of telemetry of every object around you and you should base the vehicle actions on that data.
With V2V, you are introducing unnecessary complexity to an already complex environment. That is never a winning combination.
 
The big thing that needs to be solved is perception. Pair that with good driving policy and you have a winner.

Every car already communicates - by being where it is and going as it is going, and if you eliminate human drivers you have visual indicators, a la turn signals, brake lights, etc.
With perception, you will have excellent accuracy of telemetry of every object around you and you should base the vehicle actions on that data.
With V2V, you are introducing unnecessary complexity to an already complex environment. That is never a winning combination.
Lol.

V2V (and V2N) isn’t something different or competing with FSD. It’s another communication and sensory used in FSD, one that is less complicated than human-machine as it currently stands. No has proposed it to be the only

Required V2V and V2N also allows for more efficient and effective road utilization, which not only has economic impacts, improves safety by reducing circumstances that test FSD. It facilitates planning vs reaction. Think about the roundabout problem. If all cars were FSD, communicating with each other and the network, the timing of cars utilizing the roundabout is optimized, with reduced acceleration gradients, reduced situations that become complex.
 
  • Disagree
Reactions: mikes_fsd
That's exactly why I'm disappointed with Tesla somewhat... they sold me FSD but I doubt it will be true FSD in the next 5-10 years. There is a lot required to add human instinct to computers. I'm not sure that the whole training ML concept is really going to work but in absence of true FSD, I'll take highway-capable L3 where I don't need to apply light touch every 30 seconds or so. Do little things but do it with confidence and accuracy.
 
  • Like
Reactions: DanCar
If lidar had a failure, Waymo car also has plenty of other sensors so it would probably be able to keep going. Waymo has 29 HD cameras for redundant 360 degree coverage, a main 360 degree lidar, perimeter lidar all around and perimeter radar all around! So if the main roof lidar failed, Waymo would still have 29 cameras and perimeter lidar and perimeter radar. So I think it could still drive just fine. That's the whole point of redundancy. Waymo can completely self-drive even if lidar fails or a camera fails because of the extra sensors. It would probably take multiple sensor failures at the same time to cripple the self-driving. Although it is possible that Waymo's safety procedure would still dictate slowing down out of an abundance of caution.


@S3XY CARS Just to add my last post, here is a quick little video on Waymo's new 5th Gen FSD hardware. If you pause the video, you can get some good views of the different sensors. You can see that Waymo has large HD cameras for the front and 360 degree cameras in the roof pod just below the main roof 360 degree lidar. you can also see additional perimeter cameras, lidar and radar in the front fenders, in the back and on the sides. So I think the cars would be able to self-drive just fine even with a lidar failure. And the sensors are self-cleaning and self-heating to prevent dirt obstruction or freezing in the winter. Like I said, I think you would need pretty catastrophic sensor failures all at once to cripple the FSD. So yeah, the multiple sensors make Waymo's FSD very robust against sensor failures.

@diplomat33 My apologies, but I feel you misunderstood my statement regarding a hardware failure. Yes, Waymo has multiple sensor systems, which allow it to have more redundancy than a Tesla. I wasn’t stating a failure would result in a Waymo being unable to self-drive, only that in the occurrence of a failure on any self-driving vehicle, a procedure or system must be in place regarding 'doing what is most safe for the occupants and other drivers on the road'. Of course, a single failure on a Waymo vehicle may still be completely operational, which is definitely another benefit of having multiple sensor systems, and a huge benefit to having redundancies in place; however, in the case there is a catastrophic failure of a Waymo vehicles sensors (or any self-driving vehicle for that matter), there must be a process in place to do what is best to keep the occupants and other drivers as safe as possible. In this situation, I’m stating that the question regarding ‘What would a Tesla do if the cameras, radar, sensors, etc failed?’, is valid for every company pursuing self-driving vehicles and not a one-sided argument against Tesla, as a previous user had mentioned. Every company must have a process in place regarding this situation, no matter how many redundancies or sensor systems. If one would like to argue on the number of redundancies, that’s a different matter, and of course more are normally safer, and Waymo would have greater benefit for this.

Additionally, regarding your comment on Waymo’s 5th Gen FSD hardware; are you referring to the vehicle would be fine even with an ‘entire’ lidar failure, or just a single lidar unit failure?

If you are referring to a single unit, then yes there would likely be a backup lidar unit to represent the missing data in that area, and the vehicle would likely continue, possibly at a more cautious level as you mentioned. However, if we are referring to an entire lidar system failure, I don’t see this being true, especially since we’re all in agreement that no one has a Level 4 vision-only system on the road at this time. I could be wrong, but I personally don't see Waymo attempting to solve the self-driving problem via vision-only at this time (and for good reason, since Lidar is currently proven more efficient and accurate than a vision-only system). As I've mentioned before, Tesla and Waymo are working to solve the same problem, in 2 different methods. But yes, if you were referring to a single lidar failure, then I would agree that it should still be able to drive with the remaining hardware on the 5th Gen FSD hardware you mentioned.





I agree with @diplomat33 on “Anytime Soon” is “less than 1 year” for me too. That being said, I don’t see Level 4 for Tesla in this timeline. I see a progressing Level 3 over the next 8-10 months, and a confident version with a potential Level 4 case to present to Governments in 12-20 months. I also agree with @mspisars that Tesla will be able to achieve Level 5 much quicker than its past and current progress, because of the process they’re using, and the estimated amount of data that will be received from the fleet on the road. After Tesla submits the data and information for Government/Regulatory approval, I anticipate possibly a year or more for approvals based on locations and Governments in place. Bonus if I’m wrong and it’s earlier! :)
 
  • Like
Reactions: mikes_fsd
[ After Tesla submits the data and information for Government/Regulatory approval, I anticipate possibly a year or more for approvals based on locations and Governments in place. Bonus if I’m wrong and it’s earlier! :)



Car automation legality is regulated state by state in the US anyway.

FSD (up to and including level 5- whoever gets it working) is already legal in many US states. Just nobody has such a vehicle. There's no "regulation" to get approved- it's already there.

There's basically nothing at the federal level either way on the topic other than some "suggestions" with no force of law.


The EU is a bit different, with SOME EU-wide regs, but also still additional ones in different countries....and they're somewhat draconian even regarding L2 systems (Tesla has already had to reduce functionality of EXISTING safer-than-humans systems so I dunno that any amount of data on more complex ones will change things there)

Can't speak to regulations in China or Australia, honestly haven't looked into it much.
 
  • Like
Reactions: DanCar
Car automation legality is regulated state by state in the US anyway.

FSD (up to and including level 5- whoever gets it working) is already legal in many US states. Just nobody has such a vehicle. There's no "regulation" to get approved- it's already there.

There's basically nothing at the federal level either way on the topic other than some "suggestions" with no force of law.


The EU is a bit different, with SOME EU-wide regs, but also still additional ones in different countries....and they're somewhat draconian even regarding L2 systems (Tesla has already had to reduce functionality of EXISTING safer-than-humans systems so I dunno that any amount of data on more complex ones will change things there)

Can't speak to regulations in China or Australia, honestly haven't looked into it much.

Great breakdown. Thanks for this. You're right, I was referring mostly on a global level, especially how the EU will handle things since they have so many invisible borders between the countries. I agree that European approval will be difficult for a few countries, especially those that may want to prevent Tesla from impeding in their markets. Also fair to note, I believe most European countries have safer drivers than North America. Safer and less drivers, since they have better public transit systems throughout.
As things progress, it could be interesting for road trips having Level 5 in a specific province/state or country, only to have to take over and return to Traffic Aware Cruise Control and Lane Assist before you cross the line. I think most places will jump on board around that 1 year mark after Tesla is able to submit its data and statistics. It might take 1 or 2 submissions, but the more that approve it, the quicker it will likely be adopted in nearby and similar countries.
 
  • Like
Reactions: Knightshade
Just to be super pedantic, L5 that only works in specific places isn't level 5, it's level 4.... that's pretty much the difference... L4 is no human needed, but only within specific operational domains, which can include conditions, locations, or other boundaries... L5 is no human needed anywhere ever.

Personally I don't think you can ever get L5 with the existing sensor suite Tesla is using, too little redundancy and poor performance in bad weather for example....(as discussed previously and elsewhere even the fancy Waymo taxis still have issues in bad weather and have a LOT more sensors) but I think L3 on at least highways should be pretty easy, and maybe L4 in certain conditions/locations.
 
@diplomat33 My apologies, but I feel you misunderstood my statement regarding a hardware failure. Yes, Waymo has multiple sensor systems, which allow it to have more redundancy than a Tesla. I wasn’t stating a failure would result in a Waymo being unable to self-drive, only that in the occurrence of a failure on any self-driving vehicle, a procedure or system must be in place regarding 'doing what is most safe for the occupants and other drivers on the road'. Of course, a single failure on a Waymo vehicle may still be completely operational, which is definitely another benefit of having multiple sensor systems, and a huge benefit to having redundancies in place; however, in the case there is a catastrophic failure of a Waymo vehicles sensors (or any self-driving vehicle for that matter), there must be a process in place to do what is best to keep the occupants and other drivers as safe as possible. In this situation, I’m stating that the question regarding ‘What would a Tesla do if the cameras, radar, sensors, etc failed?’, is valid for every company pursuing self-driving vehicles and not a one-sided argument against Tesla, as a previous user had mentioned. Every company must have a process in place regarding this situation, no matter how many redundancies or sensor systems. If one would like to argue on the number of redundancies, that’s a different matter, and of course more are normally safer, and Waymo would have greater benefit for this.

Yes, L4-L5 autonomous cars need to have a driving policy for handling failures regardless of what sensors they have. In fact, Waymo robotaxis can automatically pull over without human intervention in case of failures. That's the big difference between L3 and L4. L3 means that the car can self-drive but cannot handle failures (hence why they still need a human driver). L4 is self-driving where the car can also handle failures on its own without human input.

Additionally, regarding your comment on Waymo’s 5th Gen FSD hardware; are you referring to the vehicle would be fine even with an ‘entire’ lidar failure, or just a single lidar unit failure?

If you are referring to a single unit, then yes there would likely be a backup lidar unit to represent the missing data in that area, and the vehicle would likely continue, possibly at a more cautious level as you mentioned. However, if we are referring to an entire lidar system failure, I don’t see this being true, especially since we’re all in agreement that no one has a Level 4 vision-only system on the road at this time. I could be wrong, but I personally don't see Waymo attempting to solve the self-driving problem via vision-only at this time (and for good reason, since Lidar is currently proven more efficient and accurate than a vision-only system). As I've mentioned before, Tesla and Waymo are working to solve the same problem, in 2 different methods. But yes, if you were referring to a single lidar failure, then I would agree that it should still be able to drive with the remaining hardware on the 5th Gen FSD hardware you mentioned.

Both. Waymo can self-drive if just one lidar fails or if all lidar fails. Although, I suspect the odds of all lidar failing at the same time would be infinitesimally small.

Waymo has camera vision that can identify all objects. Waymo also has camera vision that can read signs, read hand gestures and know the direction a pedestrian is facing to predict where they are going to move. So I think they could do self-driving with just cameras. Here is some more info on Waymo's 5th generation cameras:

"Our latest long range cameras and 360 vision system now see much farther than before, allowing us to identify important details like pedestrians and stop signs greater than 500 meters away. Through advanced design innovations, including custom lenses and precise optomechanical engineering, our vision systems enable much higher performance levels than cameras on cars today.

In addition, our new perimeter vision system works in conjunction with our perimeter lidars to give the Waymo Driver another perspective of objects close to the vehicle. For example, while our perimeter lidars detect obstacles directly in front of the vehicle with precision, our perimeter cameras provide our machine learning algorithms additional details to reliably identify objects, providing more context to the traffic scene.

Concurrently, our new peripheral vision system helps us reduce blind spots caused by parked cars or large vehicles. These peripheral cameras enable us to peek around a truck driving in front of us, seeing if we can safely overtake it or if we should wait. Together, these various types of cameras allow us to make decisions earlier, faster, and with even more information than we've ever had before."
Waypoint - The official Waymo blog: Introducing the 5th-generation Waymo Driver: Informed by experience, designed for scale, engineered to tackle more environments

Now, you might be thinking, if Waymo has camera vision that is good enough for FSD, why don't they pursue camera-only FSD? The short answer is because camera + radar + lidar makes FSD better and more reliable. Camera-only would not be as reliable.

Camera vision is required for FSD but camera vision will never be 100% reliable in all conditions. This is what the "bible" of autonomous driving, Safety First for Automated Driving says on page 47:

"As of today, a single sensor is not capable of simultaneously providing reliable and precise detection, classifications, measurements, and robustness to adverse conditions."

Note the word "simultaneously". Yes, camera vison can provide detection, classification and measurements. So you can do self-driving with just cameras. But cameras will not always do all 3 reliably enough at the same time and in all conditions. So you might be in a situation where cameras are 100% reliable at one thing like classification but not reliable enough at another thing, like measuring distance. By combining different sensors, cameras, radar and lidar, with different pros and cons, the sum will be greater than the part. The total combined sensors will be able to simultaneously provide reliable and precise detection, classification and measurements in a wide range of conditions.

It's basically a question of how reliable or robust you need your automated driving system to be. If you just want L2 or L3, camera-only is fine. And if you don't care if the car cannot handle all conditions, then camera-only is fine. So depending on the ODD, some L4 might even be ok with camera-only. But if you want L5, camera-only is not good enough because it won't be reliable in all conditions. And if you want a super reliable and super robust FSD then you need to combine cameras with radar and lidar.

Remember that Waymo is aiming for FSD with no steering wheel or pedals. That requires the highest reliability. In a robotaxi with no steering wheel or pedals, a human will not be able to take over. And yes, L4 robotaxis can pull over if something goes wrong but you want your robotaxi to be able to self-drive as long as possible before it needs to pull over. And you want your robotaxi to be as safe as possible. So it makes sense to add extra sensors to give the robotaxi the most robust perception possible so that the robotaxi can safely handle situations on its own for as long as possible. That's why Waymo has so many sensors, not because their camera vision can't do self-driving.

I hope that makes sense.
 
Last edited:
That is not at all how you portray it in your reply.
You state is as a matter of fact, not opinion.

Thanks for clarifying that it is - after all - just your opinion.

I was always clear that it was my opinion.

I wrote "So I think the cars would be able to self-drive just fine even with a lidar failure."

"I think" indicates that it is my opinion.

The other poster asked me to clarify if I meant 1 lidar failure or multiple lidar failures. So I clarified my opinion.

But again, it is an opinion that is based on what we know about Waymo's camera vision. It is based on good evidence. It is not just some personal opinion pulled out of thin air.
 
I was always clear that it was my opinion.

I wrote "So I think the cars would be able to self-drive just fine even with a lidar failure."

"I think" indicates that it is my opinion.

The other poster asked me to clarify if I meant 1 lidar failure or multiple lidar failures. So I clarified my opinion.

But again, it is an opinion that is based on what we know about Waymo's camera vision. It is based on good evidence. It is not just some personal opinion pulled out of thin air.
Nice deflection.

Try again.

First, you state...
Waymo can self-drive if just one lidar fails or if all lidar fails.
then further down you add.
So I think they could do self-driving with just cameras.

One portrays it as a fact without a source.
The second states an opinion.
 
  • Disagree
Reactions: BlindPass
Nice deflection.

Try again.

First, you state...

then further down you add.

One portrays it as a fact without a source.
The second states an opinion.

No. Your are completely flipping the order.

Look again.

So I think the cars would be able to self-drive just fine even with a lidar failure.

That was in post #264.

Both. Waymo can self-drive if just one lidar fails or if all lidar fails.

That was in post #311. So it came after, not before.

"Waymo can self-drive if all lidar fails" came after "I think...". It was a clarification to my earlier opinion.
 
  • Disagree
Reactions: mikes_fsd
Waymo is smart and has lots of resources. I would expect them to have a camera first approach, perhaps in addition to a lidar first approach.
That is Mobileye's approach. They plan on making two independent self driving systems, one using vision and one using LIDAR. Then they're hoping to achieve a low failure rate with each one and combine them to achieve an extremely low failure rate. It's not clear to me why they think failures will be uncorrelated and how they will resolve conflicts.
 
No. Your are completely flipping the order.

Look again.



That was in post #264.



That was in post #311. So it came after, not before.

"Waymo can self-drive if all lidar fails" came after "I think...". It was a clarification to my earlier opinion.
I did... all quotes are from same post.
upload_2020-9-7_13-59-41.png


FYI, the little arrow next to an actual quote will take you directly to the post it was quoted from.
upload_2020-9-7_14-7-51.png
 
Last edited: