I wonder what the C-suite discussions were like that led to a collaboration between rivals on this topic.
-They are being hit on two major fronts simultaneously from the valley
-Cost of ML/AI experts is too high for these MBAs to fathom
-Risk of Waymo winning is existential
-Risk of Tesla winning hurts their high margin vehicle sales
my take is that the technology is too valuable (and disruptive) to not invest at all, but so expensive and far away that it doesn't make sense for an established manufacturer to give it the full court press.
If anything, I think it's the opposite. I think they both see production ready AI and being almost here, and they don't have time to figure it out themselves.
This makes joint ventures a win/win as they reduce the risk of being left behind while significantly reducing R&D costs.
(please Tesla/Elon fans don't downvote an opinion without letting me know why you thing I am wrong and start a discussions, most of the downvotes I get are on my comments about Tesla's autopilot and Apple's hardware )
The major issue is people depending too much on the system. I think it's great that the 'autopilot nag' is much harsher than in the US (except for low speeds, in stop and go, it's flawless and nags too much here). The static object thing is an issue, yes, but it's massively overblown by people who tend not to drive a Tesla, because the driver can easily spot them (if they're paying attention) and apply corrective measures. I say this, because as of right now, Autopilot is a driving aid, not a replacement. (Yes, people have something to say about marketing, but that's a different discussion.)
To me, the biggest issues with Autopilot (as it is supposed to be used today), are phantom braking and unexpected steering inputs. These require instant correction, are completely unpredictable and have the ability to scare other drivers or even cause crashes. Also, they happen a lot. Finding static objects on a highway is much less common than just driving on a straight road while being tailgated and therefore the impact is much bigger.
Another thing is Autopilot unexpectedly disabling and the driver not being aware, because the indicator is not that clear. Sometimes it's a pleasant sound and if you don't look at the screen because you pay attention to the road, you might miss the visual indicator.
Outside of the US, the major issue is that it's being built for the US and isn't fully adapted to non-US roads and rules. The 'Emergency Lane Departure Avoidance' feature (enabled by default before every drive) caused massive issues in the Netherlands on roads that have shared bike lanes. Also, in the Netherlands we have 'rush hour lanes', which are actually the hard shoulder, which is temporarily designated as a lane, as indicated by a green arrow over the lane. Autopilot doesn't know this and wants to move over, because it thinks the lane ends.
To get back to your point, static object recognition is a small thing to overcome and is, so it seems to me, massively inflated by media that are heavily opinionated against Tesla. There's a much bigger problem in getting it working for every country and every weird situation. And there are things that humans do that I expect will be hard to pick up in an automated system. For example: spotting the driver next to you playing on the phone and passing for safety, seeing headlights reflected in a window and knowing a car is about to round a corner, etc.
> phantom braking and unexpected steering inputs. These require instant correction, are completely unpredictable and have the ability to scare other drivers or even cause crashes.
That sounds kinda scary to me. I'd rather drive a car that doesn't try to do anything smart than one with these issues.
> Also, they happen a lot.
Wow. If that's your perception as a Tesla fan, it doesn't exactly inspire me with fresh confidence.
Also, my previous car, an Opel, did the same thing because of a specific type of line marking messing with the emergency braking system. That was reproducible, but the car doesn't phone home or get updates, so it the same thing happen each time I passed a specific road. At least I know the Tesla will get better.
Why do you think so? There will be often enough stopped cars on the road to cause issues, or cement barriers, construction sites.
Do you think that having a filter that removes the objects with speed== 0 from the AI is something an safety engineer should do ? For me it looks like optimizations game developers do when your video card/CPU is not powerful enough so small objects are ignored at a larger distance.
Driver issues are a different topic, since they signed that TOS they will get all the blame for killing themselves.
IMO true self driving(without a human driver) will need more powerful hardware so no safety compromising shortcuts/optimizations are done(not even 1)
Yes, but it's one big thing. Fixing one big thing, usually, is much easier than fixing thousands of little things. My point wasn't that it's not important, it's that it's not the biggest problem, even though the (mostly uninformed) media and keyboard warriors are making it out to be, which is distracting from the actual issues.
> IMO true self driving(without a human driver) will need more powerful hardware so no safety compromising shortcuts/optimizations are done(not even 1)
Hardware 3 has only just been released and is apparently at least 10x more powerful. Give it a chance for the next couple months, anything is possible and we have to wait and see.
I agree, but in this case may mean wait for the hardware to be powerful enough to be able to handle all the inputs in real time.
>is distracting from the actual issues.
What hardware/software more important issues are you referring ? Non related to that the bigger issues is the hype IMO, I only seen bad sentiments for Tesla and uber but not for the others that self driving companies.
I personally really want self driving to be ready as soon as possible, I can't drive due to medical issues, I live about 10Km away from the city and having the mobility of a car would help me a lot(I am not exaggerating)
No robo taxis in 2020 then.
Being what I would consider a beta tester for self driving cars seems a silly risk. But, people also pay lots of money driving vintage cars without modern safety systems which seems insane to me.
It's the same story as Apple in its heydays (the two Jobs eras): they radically altered the markets for PCs, MP3 players, and smartphones, largely by realizing that they had the ability to do so.
Here's the items that need to be overcome:
1) For consumers that don't have a garage, they have to use public charging infrastructure.
2) For those who charge at home, if they do any extended driving they need to use public charging infrastructure
3) The current public charging infrastructure doesn't charge fast enough. For case 2, you can get by (sometimes) with charging at 150 miles in half an hour. Not acceptable for case 1.
4) Most importantly, battery wear and cost. Let's say a battery that costs 30K, for 300 miles range, lasts 10 years. Cost of a "fill up" (300 miles) is about $12.00 of electricity, and $60.00 of wear on the battery. Yes, the battery isn't totally worn down, but they are less useful for automotive use if the capacity is significantly lessened.
Now the larger the battery size, you can get by with charging to 80% capacity, and discharging to about 15%. This still significantly increase the longevity, but the cost structure is such that 200 mile batteries (or lower) is all you find in "affordable" electric vehicles. So for this to really take off, I believe you need a battery capacity of 500 miles, at a cost of 15k, with charging time of 300 miles range in 15 minutes. (Tesla is there on the charging time with their one example of a "1000 mph" charger, so hopefully this will be rolled out to the wider public across more manufacturers).
Battery costs have been dropping rapidly, %35 in the last year alone. Starting around 2023 they will be low enough that the sticker price for an EV will be about equal to a comparable ICE car.
Given that EV operating costs are considerably lower, and that governments around the world are pushing EV's, sales of them are going to skyrocket. The big manufacturers have recently realized this, which is why they are switching their EV programs into high gear.
If self driving cars are only equal to good human drivers in risk it is could be a fad.
Also, PR is pretty sensitive for automated cars. They can't just be statistically equal to human drivers, they have to be at least an order of magnitude better and they cannot afford to make any obvious errors that a sober human would not. Especially if it results in a fatality.
Note that the ban isn't on all cars on the road, it is all new cars. 15-20 years latter it will come to all cars when human driven cars are mostly collectors items anyway. Until then self driving cars will need to deal with humans.
Probably "nobody cares about the technology anyway; we'll differentiate on branding, like always. A car is a car."