For level 5 Autonomous driving (driving without anyone in the drivers seat or in the car) Elon Musk has previously stated that the system needs to have redundancy on all levels. I can not see that this system has the required redundancy to achieve this. For instance there is only one computer doing all the calculations, there seems to be gaps in the redundancy of the cameras (what happens if you get dirt on one of the front or rear facing side cameras? Elon Mentioned that the car will alert the user, but what if there is no one in the car?), the Ultrasonic sensors seems to have redundancy except for on the sides of the car, there is only one radar. This systems seems, to me, to be only able to achieve level 4 of autonomous driving due to its lack of Hardware. My guess would be that this is only V1.0 of Teslas take on Autonomous driving and that for true level 5 Autonomy they would need to further upgrade this Hardware suite. Anyone have any thoughts on the subject?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Can level 5 Autonomy be achieved with Hardware suite 2.0?
- Thread starter Øivind Hoem
- Start date
byan1232
Member
Even if Teslas can be "self-driving", I would not use it fully. Autopilot is perfect the way it is as a driver assist.
There are situations on the road where human decision making must happen. For example:
-Pothole (we can see the depth and determine whether to steer clear or not)
-Objects on the highway (ever notice a big metal piece, possibly part of someone's bumper, in the middle of the highway?)
-Little kid running across a local street
There are situations on the road where human decision making must happen. For example:
-Pothole (we can see the depth and determine whether to steer clear or not)
-Objects on the highway (ever notice a big metal piece, possibly part of someone's bumper, in the middle of the highway?)
-Little kid running across a local street
The world is not quite ready to answer the L5 autonomous regulations, standards and approvals. I suspect by the time there are some regulations in place AP 3.0 will be released to fully satisfy those requirements.
Until then, like byan1232 states, it would be an awesome driver assist.
Until then, like byan1232 states, it would be an awesome driver assist.
How do you know that they didn't include a redundant computer/GPU?
It is possible that they priced Enhanced AP and Full Autonomy high enough to include the option for them to retro-fit additional sensors on the car in 1-2 years when L5 software is ready if it becomes necessary. (The sensors will likely be cheaper then, and they would serve little purpose now.)
Hopefully someone will tear their car apart and see if there are extra wiring harness connectors around for things like adding a rear facing radar.
It is possible that they priced Enhanced AP and Full Autonomy high enough to include the option for them to retro-fit additional sensors on the car in 1-2 years when L5 software is ready if it becomes necessary. (The sensors will likely be cheaper then, and they would serve little purpose now.)
Hopefully someone will tear their car apart and see if there are extra wiring harness connectors around for things like adding a rear facing radar.
essmd
Phantom X
I agree with the original post, since it may very well be that the closer full autonomy becomes a reality, so may be the recognition that AP 2.0 won't fully cut it. One element used in other autonomous vehicles is LIDAR, which provides an image of the surrounding environment very differently than just visual, sonar, and radar data acquisition provides. It could be 3.0 will be required to achieve the goal.
No, I'm fully convinced that L5 driving cannot happen without a significant number of upgrades to our infrastructure.
Plus we have to ask ourselves what the point is in upgrading a car without upgrading the road? we have hugely inefficient stop light systems for example.
Around where I live they are not synced, they are not intelligent. They just randomly create traffic jams because they can't react to the traffic flow.
What I see happening is bringing roadways and cars up the level 5 driving. Where there will be designated roads that are white listed for L5 driving.
Once people realize how much more efficient this system is more and more roads will follow.
It's likely going to require addition hardware (for road to car communication, and car to car communication) than what the Ap 2.0 hardware has. It's also going to require a lot of regulations to be written. Plus we have to find a way to get bicyclist to wear transponders to insure that they are always recognized/seen.
Plus we have to ask ourselves what the point is in upgrading a car without upgrading the road? we have hugely inefficient stop light systems for example.
Around where I live they are not synced, they are not intelligent. They just randomly create traffic jams because they can't react to the traffic flow.
What I see happening is bringing roadways and cars up the level 5 driving. Where there will be designated roads that are white listed for L5 driving.
Once people realize how much more efficient this system is more and more roads will follow.
It's likely going to require addition hardware (for road to car communication, and car to car communication) than what the Ap 2.0 hardware has. It's also going to require a lot of regulations to be written. Plus we have to find a way to get bicyclist to wear transponders to insure that they are always recognized/seen.
If the "infrastructure" needs to change to accommodate the shortcomings of a so-called "level 5" vehicle, then by definition the vehicle is not "level 5". Folks, it's really as simple as that. This is getting silly!
The infrastructure has to change all the time to accommodate the short-comings of a human, and a human is level-5.
The infrastructure has to change all the time to accommodate the short-comings of a human, and a human is level-5.
Haha, first time I got a dislike on a joke.
Tam
Well-Known Member
For sensor redundancy, there's a trifocal camera in front as opposed to the previous classic single monocular camera.
If the trifocal camera fails, the front radar and sonars can take over to bring the car to an emergency abort so a tow truck can take it to a service center and fix the problem.
Each nVidia Drive PX 2 motherboard has 2 different combined GPU+CPU chips (Graphic Central Processing Unit + Central Processing Unit.) So if one combined GPU+CPU chip fails, the other one can take over to activate an emergency abort procedure.
For Autocruise, you only need 1 GPU for each motherboard.
For AutoChauffeur, you need 2 GPUs for each motherboard.
For Self-Driving, you need "multiple" boards.
So if 1 board fails, the other board can activate an emergency abort procedure as well.
I call it as motherboard, but each one of these is actually a supercomputer as Elon call it.
But as usual, Tesla keeps finding ways to refine the car so I will not be surprised if this is not the end of any hardware change.
Last edited:
I wouldn't argue against the utility of "whitelisting" regions where or conditions when the car is most likely to function properly. But we can't also refer to such a system of caveats as being "fully autonomous". The industry will wind up needing to qualify what "fully autonomous" actually means compared to the sensibilities of a human driver.
How about a scenario where the destination address happens to be a building on fire. Would the "fully autonomous" car find a nice spot next to a blaze or would the AP heat sensors detect the danger? Can the car recognize a road closing (police, crowd, event, etc) several blocks away and plan a detour? How can the car tell the difference between a cop in the middle of the road directing traffic and a cop telling it to go back the other way? For that matter, how can it even understand a cop directing traffic? How would it handle a live power line draped across a parallel parking spot?
How about a scenario where the destination address happens to be a building on fire. Would the "fully autonomous" car find a nice spot next to a blaze or would the AP heat sensors detect the danger? Can the car recognize a road closing (police, crowd, event, etc) several blocks away and plan a detour? How can the car tell the difference between a cop in the middle of the road directing traffic and a cop telling it to go back the other way? For that matter, how can it even understand a cop directing traffic? How would it handle a live power line draped across a parallel parking spot?
Johan
Ex got M3 in the divorce, waiting for EU Model Y!
If the current hardware (being put in to cars built from now and going forward) does not in fact suffice to create true level 5 autonomy (which may very well turn out to be the case) then Tesla have straight out lied. It wouldn't be the first time that they've overpromised.
The levels are vaguely worded. People can and will accept 90% autonomous capability as being the same as "fully autonomous".
The many incremental changes provided above that 90% will provide "enhanced" full autonomy, thereby making it seem a bonus.
Tam
Well-Known Member
When talking to the public, best scenarios are usually described.
Elon promised that AP1 could be handfree from freeway ramp-to-ramp. It will still happen but with one small footnote that you still have to babysit the AP1 and be ready take over at all times.
Elon promised that this time, L5 will be achieved and Tesla will demontrate a driverless trip from LAX to NYC by the end of next year.
I believe that demo will be done. What we don't know is: Are there any footnotes?
Tam
Well-Known Member
What Tesla describes seems not to be vague! You can start your driverless ride hailing service via Tesla Network.
So L5 is real.
What I understand is: L5 does not mean that all accidents, injuries and deaths will be zero.
Autonomous Driverless cars can reduce them but not eliminate them.
Tam
Well-Known Member
Regulatory process is only a minor inconvenience for now.
For now, we need to know how reliable the system is.
While the system is being trained, of course, there would be mishaps.
But I hope, after that phase, the system can handle almost all scenarios without getting human into troubles.
Once, the system will achieve L5 competency but the law forbids L5, then don't drive the car without a driver.
A driver can just sit there and enjoy the ride until the law changes!
This is actually a two-part question:
1. Will this hardware reach L5 capability?
2 Will it then be L5 approved (by regulators)?
For #1, it should be noted that Tesla is one of the few car companies attempting L5 with only radar and cameras. Nearly all other companies are using cameras and LiDAR. Thus, being the (only) one trying for a cheaper solution is not without risk. It's possible L5 is unachievable without further hardware additions. It's by no means a guarantee that Tesla achieves L5 with their current sensor suite and hardware.
For #2, the regulations don't yet exist to describe what specific redundancies will be necessary. This is another risk. Will they require multiple processors, sensors, braking systems, power supplies? We don't know.
Thus, there's essentially one way to get it right, and many many ways to come up short by even one requirement or one corner case.
All in all, I would guess the odds are very good that this hardware reaches L4, but probably less than 50% that it becomes L5 approved.
1. Will this hardware reach L5 capability?
2 Will it then be L5 approved (by regulators)?
For #1, it should be noted that Tesla is one of the few car companies attempting L5 with only radar and cameras. Nearly all other companies are using cameras and LiDAR. Thus, being the (only) one trying for a cheaper solution is not without risk. It's possible L5 is unachievable without further hardware additions. It's by no means a guarantee that Tesla achieves L5 with their current sensor suite and hardware.
For #2, the regulations don't yet exist to describe what specific redundancies will be necessary. This is another risk. Will they require multiple processors, sensors, braking systems, power supplies? We don't know.
Thus, there's essentially one way to get it right, and many many ways to come up short by even one requirement or one corner case.
All in all, I would guess the odds are very good that this hardware reaches L4, but probably less than 50% that it becomes L5 approved.
Similar threads
- Article
