Apple abandons self driving car/says it’s not technologically feasible to achieve L5

[diplomat33]

Really speaking, L5 cannot and should not mean it can replicate the best ever human driver.

It does not. SAE defines the L5 ODD as that of a "typical human driver", not "the best ever human driver":

“Unconditional/not ODD-specific” means that the ADS can operate the vehicle on-road anywhere within its region of the world and under all road conditions in which a conventional vehicle can be reasonably operated by a typically skilled human driver. This means, for example, that there are no design-based weather, time-of-day, or geographical restrictions on where and when the ADS can operate the vehicle. However, there may be conditions not manageable by a driver in which the ADS would also be unable to complete a given trip (e.g., white-out snow storm, flooded roads, glare ice, etc.) until or unless the adverse conditions clear. At the onset of such unmanageable conditions the ADS would perform the DDT fallback to achieve a minimal risk condition (e.g., by pulling over to the side of the road and waiting for the conditions to change)." (p 32)

"Level 5 is distinguished from Level 4 by the fact that it is not operationally limited to a specific operational design domain and can rather operate on-road anywhere that a typically skilled human driver can reasonably operate a conventional vehicle." (p 38)

Yes, L5 would need to pull over when conditions are not ok for its sensors and computer. However, there is still the expectation that L5 can handle any condition that a typical human driver could handle. So it would be ok if the L5 sensors and computer pull over because they cannot handle a total blizzard or flooded road but it would not be ok if the L5 sensors and computer have to pull over because they can't handle light rain.

 

[diplomat33]

I would add that AVs should be safer than the average human driver. After all, if they weren't, there would be no reason or benefit to deploy AVs. But safer than the average human driver does not mean that AVs have to be 100% perfect. There is also the matter of at-fault accidents. We certainly don't want AVs to cause accidents on purpose. But there will be instances where the human driver is at-fault and the AV could not realistically avoid the accident. We should not blame AVs for that.

 

[enemji]

However, there is still the expectation that L5 can handle any condition that a typical human driver could handle. So it would be ok if the L5 sensors and computer pull over because they cannot handle a total blizzard or flooded road but it would not be ok if the L5 sensors and computer have to pull over because they can't handle light rain.

Definition of the average human driver is still subjective. Everyone I have seen complain appears to hate self driving based upon their own strengths and weaknesses. They expect their strengths to be represented as well as their weaknesses to be not represented/reflected in the self driving experience. That is not an achievable objective.

 

[voyager]

L4-L5 when you're cruising from LA to Las Vegas, all traffic on your side of the divider going in the same direction.
L3-L4 at best in urban traffic.

 

[diplomat33]

Definition of the average human driver is still subjective. Everyone I have seen complain appears to hate self driving based upon their own strengths and weaknesses. They expect their strengths to be represented as well as their weaknesses to be not represented/reflected in the self driving experience. That is not an achievable objective.

When talking about the average human driver, we are talking about the average accident rate, ie total accidents divided by total miles driven. We can certainly achieve AV safety that is far better than that average accident rate.

And AVs will never be 100% perfect but it is absolutely achievable to get AVs that are better than human drivers. IMO, AVs are already superior to human drivers in many areas. AVs have better perception that humans. AVs don't have blind spots like humans do. AV perception does not degrade with age like with humans. AVs have better reaction times than humans. And AV reaction times are always optimal and does not degrade with age or due to distraction or some other impairment like with human drivers. AVs can also make good predictions for hundreds of other objects around then much quicker than human drivers. Lastly, AVs don't make emotional decisions when driving like humans can. For example, you will never see an AV have a road rage incident.

The only area where AVs lag behind human drivers is that humans are better at solving a new situation that they've never seen before. AVs can sometimes encounter a new situation and not be sure what to do or do something stupid that a human driver would never do. But this problem can be mitigated with better ML, more data and more training.

 

[enemji]

When talking about the average human driver, we are talking about the average accident rate, ie total accidents divided by total miles driven. We can certainly achieve AV safety that is far better than that average accident rate

That goes back to my earlier comment that if the vehicle stops due to a limitation of its capabilities and restarts when that is no longer applicable, all in the goal of safety, it is achievable but nobody will like it since it will not align to a human’s capabilities and expectations for that drive. For example people complain about not doing a rolling stop for crying out loud. Tesla introduced it due to customer demand but it is stupid af to do a rolling stop. It has to be a binary. Either you stop or you don’t. Then people also complain about the car lengths distance since human drivers kept getting in the middle.

 

[diplomat33]

That goes back to my earlier comment that if the vehicle stops due to a limitation of its capabilities and restarts when that is no longer applicable, all in the goal of safety, it is achievable but nobody will like it since it will not align to a human’s capabilities and expectations for that drive. For example people complain about not doing a rolling stop for crying out loud. Tesla introduced it due to customer demand but it is stupid af to do a rolling stop. It has to be a binary. Either you stop or you don’t. Then people also complain about the car lengths distance since human drivers kept getting in the middle.

In terms of the car stopping due to say heavy rain or snow because the AV cannot handle it, it will be important for the AV company to develop the autonomous driving to be able to handle whatever a typical human driver can handle (see L5 ODD definition that I posted earlier). If the AV only stops for severe rain or snow that humans should not drive in, then the human cannot blame the AV.

In terms of things like rolling stops, I think people will need to adjust to the fact that AVs may drive differently than they do. And the fact is that human drivers don't always drive the right way. Sometimes humans do drive too close to the car in front in an unsafe manner. So AVs should not imitate every driving habit that humans have because they are not all good. At the end of the day, AVs will never please everybody. But it is not the goal of AVs to make every human driver happy, the goal is to drive safely and reliably.

 

[enemji]

In terms of the car stopping due to say heavy rain or snow because the AV cannot handle it, it will be important for the AV company to develop the autonomous driving to be able to handle whatever a typical human driver can handle

Therein lies the problem. I don’t drive at night past midnights on a highway whereas my friends drive all night long just so that they can get to their vacation spot bright and early in the morning. For me, if the AV stops working in the night, I have no problems. Not so for my friends. So expectations needs to be reset for the adoption

 

[diplomat33]

Therein lies the problem. I don’t drive at night past midnights on a highway whereas my friends drive all night long just so that they can get to their vacation spot bright and early in the morning. For me, if the AV stops working in the night, I have no problems. Not so for my friends. So expectations needs to be reset for the adoption

What are you talking about will be clearly defined in the level of autonomy, ODD and usage intent of the AV. The manufacturer will clearly define the ODD, ie the when and where the AV can be used. Everything inside the ODD, the AV needs to always be reliable and safe. It will be up to the manufacturer to make sure that the AV always works safely and reliably inside the ODD. So if nighttime or rain or snow are inside the ODD, the AV should never stop because it can't handle it. Anything outside the ODD, the AV can stop or request the user take over. So if nighttime or rain are outside the ODD, then it is ok for the AV not to stop or ask the user to take over. The ODD should be clearly communicated to the user. The manufacturer is not going to just develop an AV and deploy it everywhere and the consumer be like "oops, I guess the AV is not very good at night driving so it stopped". The public should know all this information before they use the AV. And yes, we will need to educate the public about how AVs work. But that is true for any tech.

 

[azred]

To be clear, I don't think Apple is saying that L5 in general is not possible. They are saying they can't achieve L5. That is an important distinction to make.

But frankly, I don't know why people even pay attention to these stories about Apple's self-driving car. Maybe because Apple is a big and famous company? But the reality is that Apple's self-driving project has never seemed serious to me. Apple was never going to do L4 or L5. They only did 16k autonomous miles for all of 2021 and their disengagement rate was 1 per 20 miles. They have not released any details about their autonomous driving that I am aware of. All we have are fan concept art of an "apple car". So I am not surprised at all that Apple is rescaling to just do autonomous driving on highways. My guess is Apple will try to create a L2 "hands-free highway" or maybe "L3 highway" driving system to sell to automakers. They might even try to follow comma and develop some type of "L2 hands-free highway" kit that they could sell direct to customers. I could see that as a viable business model since Apple has millions of loyal customers who would likely buy a kit that makes their car L2 "hands-free".

Thanks to Blackberry's arrogance, lack of foresight or whatever other reasons, Apple went from an innovative personal computer company that had a good chance of going bankrupt into the cell phone god with unlimited piles of money. It is of course interesting to follow their “plans” for the next big thing to spend some of that money. Unfortunately for them, I think they need another visionary like Jobs much more than an electric car in an increasingly crowded field.