Question About Last Week's Announcement

Yes, No, maybe?

Yup, that what I understand...

No (probably). new cars can use some or all of the hardware for automation UP TO full autonomy. At the very least, 4 (3 more than before) of the cameras can be used for the advanced autopilot which is available... soon, and I suspect that we will see at least some features from the "full self driving" stuff be available even if actual full L5 self driving is delayed due to legislation.

I also suspect that all of the new hardware will be feeding into the base safety features as well, increasing how well those work.

 

Elon Musk said that until a software update in December the new cars will _lack_ existing functionality. After that they'll add new features as they as implemented and validated.

 

They cannot enable a feature such as that and not tell anyone. It must pass regulatory approval. This doesn't mean the system is not active collecting data... because it is.

Right now the hardware will have no features until dec/jan when enhanced autopilot will be enabled. Enhanced autopilot is safer than current autopilot and should also allow freeway transitions to other freeways and on ramp to off ramp autonomy.

Self driving is still many months off for that to be enabled.

 

First, lets use Tesla's terminology for clarity. "Full autonomy" is what Tesla is calling Full Self Driving Capability (for brevity lets say FSDC).

All Teslas currently being built have all the hardware needed for SFDC, but the software is not yet available and Tesla is not giving a specific date or even a timeframe for when it will be available to those who purchase that option right now.

Tesla did not say that FSCD will be available in "a year or so". Elon said that in a year his goal is to demonstrate a Tesla driving from LA to NY without the driver touching the controls and the car will Supercharge as needed without human involvement.

 

I'm laughing at myself because in a single post I managed to type the same acronym 3 different ways! Sheesh... they all should be FSDC.

I can't predict the future, certainly not with a company that innovates at the rate Tesla does.

My guess is that for some time to come Tesla will maintain two different self driving options as they are currently described on the S and X configuration pages: Enhanced Autopilot and Full Self Driving Capability. Obviously Enhanced Autopilot includes a subset of the features that FSDC will include.

 

Please remember that Elon's statements are usually rather optimistic and bold. Expect a software update every couple of months that will incrementally improve autopilot function. Sometime in the next couple of years, the car will be fairly autonomous in certain situations. Also consider that over the next year or two Tesla will upgrade the hardware installed on the cars to the point that today's hardware will be obsolete.

 

AP1.0 is "fairly autonomous" in limited situations. The following are my personal observations.

The traffic adaptive cruise control is very reliable on freeways. There are many situations where it is not necessary to use the brakes or accelerator. But AP1.0 does not read surface street features like lights, stop signs etc. AP2.0 hardware has this capability but it may be a couple of years before the software is developed far enough to enable surface street autonomy.

The Autosteer is more limited than TACC. It does not handle curved roads/parkways well; it needs Interstate Highway type gradual curves to maintain lane keeping. It reads lane markings well when on both sides but less reliably if only on one side. AP2.0 hardware has this capability and I'd guess the software will be up to this in a year or so.

AP1.0 blind spot detection is not good, so you must look before using the auto lane change. The rear side view cameras in AP2.0 should solve this problem.

 

I seriously doubt it is there now. Even replacing the current functionality using the new hardware won't be ready until Dec/Jan.

Full autonomy requires a lot more. Identifying roads/parking lots to turn into, pulling out into traffic, handling construction zones. The scope will likely expand over many many months. It isn't even 100% that the current hardware will handle it.

It would take me a long time to get comfortable with an autonomous car trying to cross two lanes of traffic on a busy highway to turn left and merge into traffic.

 

Personally I still think we're looking at years... 10 to 15 at least but likely more. Tesla is definitely going about it the right way - putting the system out there and slowly expanding the envelope of situations that the cars can handle. And the fleet learning is probably the only practical way to achieve a high level of performance.

But, the lawyers are just barely agitate the fecal matter. There are many years of fees to collect before people will be able legally lay back and have a snooze. I don't think I'll see it. My kids, maybe.

 

I think that once the tech is sufficiently good, in the next 1-5 years, it will be approved for use fairly quickly. The fact is that it is a very important issue for the economy. The cost of transporting goods and passengers can probably be halved, which will lead to a massive increase in productivity and competitiveness.

Some countries will be quick to adopt full self driving, and the rest will have to follow their lead or face serious economic consequences.

 

I suppose we will all see. But I see it taking a very very long time to deal with the extreme corner cases and the "Who do I kill?" scenarios. And the moment that the driver is legally permitted to lay back and have a snooze, those things become very very important.

I'm also concerned that a self driving car, though perhaps statistically safer than a human, may well fail in completely different situations and will therefore kill different people. It would be hard to sit and court and argue "Yes, little Jimmy is dead and no a human wouldn't have killed him. But statistically Bobby and Mary are alive.".

 

How many times have YOU had to decide between hitting two people?

The car won't find itself in that situation as easily as a human could for starters (read almost never), second the person hit will probably be closest to the visible open path. With a DNN in the exact same situation you will get reproducible results. If in a situation it hits little jimmy then it would have always hit little jimmy.

For a self driving tesla to hit a person it will have probably detected it at the last possible second in which even the fastest humans couldn't have made a decision in that time anyway, or it didn't register at all and there was no choice to be made. The third option is that the car is out of control, but therefore by definition, out of control (no decision to be made either).

 

Whether it's happened to me is irrelevant. The law of large numbers dictates that that will most definitely will happen. There will be the "Do I head-on the transport that just crossed into my line or kill the 3 kids on the sidewalk?" scenario. And there be many many others. And when it happens there will be long drawn out lawsuits with people up in arms demanding that the technology be banned.

If anybody thinks that this will go smoothly, I suggest that they've missed the whole anti-vax poop show - where the net benefit to society is light years beyond any conceivable rational dispute. Yet people dispute and both lawyers and politicians listen.

 

This is easy, there's no clear path due to obstacles. It would likely pull to the far side of the lane and slam the brakes. If the oncoming transport hits it then it hits it.

There Are no ethical or moral decisions that need to be made. It's all logic.

Again, if the transport veers over at the last instant then, if the computer registers the transport, it'll maneuver faster than you could. If it veers over further out, then the car will see the transport and act before you would. In both cases it'd result in a more positive outcome than a human driver.

 

You're missing my point. If you want me to spell it out further, imagine that the transport has pulled out to pass, there is another vehicle in the opposing lane and a guardrail beside them.

The point is that there are scenarios where there is no good way out. With large numbers they WILL happen and the car will make a decision. And that WILL lead to legal action, as it already has. It will take years to sort out.

I'm as big a fan of this technology as anybody. But our ponderous and creaky legal system will get involved. For sure.

 

The action is to brake suddenly... there are no other possible actions which could be taken by a human or computer... and the computer will brake before you could. Again a positive outcome vs human.

No matter what you come up with it will typically be the fault of another human and if a human was driving then it'd likely result in the same issue and possibly a worse outcome. There are exceptions, of course, and it's never going to be perfect, but it'll be safer than a human. We hope the press takes into account statistics when/if something happens. This also means Tesla needs to be reporting often.