The revolution is the fact that they have to compete with other car manufacturers at making electric cars.
Ten years ago the common wisdom was that electric cars were all ugly, impractical, underpowered golf carts, too expensive to make, and nobody wanted them.
Dragging all the car manufacturers into the 21st century grumbling all the way, and changing public opinion about electric cars, was the stated mission of Tesla. Even if Tesla flops as a profitable enterprise, their mission succeeds as long as electric cars become a significant percentage of vehicles driven, easing carbon emissions for decades.
Complete victory of course is if almost every manufacturer shifts to only producing electric vehicles, or just has one specialized gas model for novelty buyers.
10 years ago nobody thought about soldering up thousands of flashlight batteries to run a car.
You seem to forget that electric cars were made 100 years ago and were successful in their day. they died out when gas stations became common place and greatly extended the range of cars. The early electric cars were actually pretty good vehicles for city driving and very reliable.
If Tesla flops they’ll set the EV market back a few years. Fingers crossed they don’t pull an Oldsmobile diesel and fowl the water.
It’s on the table, but not practical to accomplish. If the metric is 72 jph, then it doesn’t matter whether any one particular unit could have been built faster, because that very last station is the bottleneck limiting you to 72 jph. Supposing that we could parallelize an operation such as the moonroof. We run non-moonroof at 73 jph, and moonroof at 71 jph, and the take rate is 50%. What do I do with the fast moving product? Do I train the employees at the final bottleneck to work at 73 jph? Or do I invest in a decoupler to let them work at 72 jph, and idle the moonroof line? What if the take rate changes? Do I layoff the affected workers?
Sure, it’s possible to optimize all of those little wastes out of the process, but you’re going to end up with waste elsewhere, or violate labor contracts.
I saw a red Model 3 on the streets of suburban Las Vegas tonight!
At first, seeing it from a distance head-on, I thought it was a Model X, because it had the same grill-less nose. Then it went past me and I saw it was a sedan and had the flush door handles with the little upward tail at the end.
I didn’t expect to see one so soon. Are there any numbers on how many are actually out in the wild now?
Agreed; it’s a hard and sometimes impossible problem. And it does depend on the particulars. If your line has 100 stages and all but one are running at 72 jph, and the last at 71, then you are going to work like hell to speed up that last one. Maybe you push the workers harder but run 4 shifts instead of 3 and intersperse lots of breaks. Maybe there are some robots which are 5% faster but twice the cost. Maybe you just spend a disproportionate time optimizing the movements.
Well, I doubt any of this is at all novel thinking. I’m sure that station balancing is an art and science all by itself.
If the stations were highly reconfigurable, then you could just have the cars circulate through a generic pool of them, and spend only as much as time they need. An optioned out version might spend 100 “cycles” in the pool; a barebones one only 70. The pool is always fully utilized because there’s no specific sunroof station; if a car doesn’t have one, it just goes to the next step. When one car leaves the pool, another enters.
Of course this is tricky because it requires full automation–a human crew isn’t going to be able to switch tasks like that. And even the individual stations are going to take time to switch, so you’d need some system for keeping the transitions to a minimum. It couldn’t be fully generic either, since there is such a disparity in the size and handling characteristics of a part; a dash is very different from a seat or a door or a radio.
Still, maybe it could make sense in some situations. It helps with the quantization problem; that if you’re right below some performance threshold for some task, then to do better with just one machine means you now need two machines that are only a tad more than 50% utilized. With a pool of 100, if you need 1% more perf then you add one more machine.
Nice! Someone at work also just got one, also in red. I’m glad I got that color.
Highest VIN spotted in the wild is 11188. They number sequentially and don’t tend to leave huge gaps, so that’s probably within a few hundred of the total.
Yeah. And I think it’s clear that Tesla, or some equivalent, was necessary. Other automakers are still dragging their feet, but I think the trend is inevitable at this point and any continued naysaying is mostly for the sake of calming shareholders until they get their ducks in order.
Still, Tesla does have to figure out their manufacturing. I think they’ll pull it off this year, which still puts them a couple years ahead of most of their competition (BMW is saying it’ll be 2020 before they really have a serious EV lineup). But it’s close.
I’m not sure how sarcastic you’re being here, but this really is no mean feat. It was the first use of lithium ion in a production car. And they managed to qualify a non-automotive product for automotive use–and not just any use, but the heart of their drivetrain. They pulled it off. The Roadster wasn’t a perfect car but the battery wasn’t the weak point.
They already do. No one can deny that Tesla is an immensely desirable brand, both in general terms and in demand for the S/X/3/Semi. AI Proofreader is right–they haven’t “won” until virtually all ground transportation is electric. I give it 20 years; 30 at the outside.
It’s a really frickin insane idea. And Elon pulled it off.
They already do. No one can deny that Tesla is an immensely desirable brand, both in general terms and in demand for the S/X/3/Semi. AI Proofreader is right–they haven’t “won” until virtually all ground transportation is electric. I give it 20 years; 30 at the outside.
[/QUOTE]
I disagree that it’s a win. The win is matching machine to task.
Electric cars may be common in 20 years but it will take a lot longer to replace all cars with electric ones.
ICE will still be used in very cold climates (ie Northern Canada), in areas that require a long range before refueling.
With the average car on the road for 12 - 15 years, it will take a good 30-40 years to replace them.
Of note I wonder if any Model T or even Model A Fords are still daily drivers. I know I guy at my church who has a driveable and licensed Model A (not a daily driver)
Yeah, I should say that I mean 20-30 years before virtually all new ground transportation is electric. As always, it takes a while to get rid of the old stuff. Though in some cases, the advantages of EVs may push even functional ICE vehicles out. This is likely to happen for trucking first.
Interesting in that you think trucking will be the first. I assume you mean heavy trucking . Will the electric trucks have the range for this. (A lot of truckers will go a long time before refueling)
It may take time for the support network ( ie charging stations) before this becomes a realty. That may be the limitations on this changeover.
Of note, the truck used in the TV show BJ and the Bear is still running and that was a 1980 model. (of course, it was restored)
The Tesla semi will go 500 miles on a charge, and can charge 400 miles in 30 min. Legally, truckers can’t drive more than 11 hours a day, nor go more than 8 hours without a 30 min break. So, a driver can go for 7 hours, or 420 miles @ 60 mph, top off 250 miles in much less than a half hour, and still have plenty for the remaining 4 hours / 240 miles. At the end point, or simply at the end of the day, the truck has plenty of time for a full charge. At no point is the truck idle when an ICE truck would not already be idle.
Of course, self-driving may put a kink in things–fully autonomous trucks don’t need rest breaks. But I think the cost savings will still exceed any potential downtime. It won’t be much, anyway; the charge time is <10% of the total.
The Tesla truck has so much power that it doesn’t need to slow down when going uphill. In some areas, it’ll be a big time savings going uphill at 60 mph instead of 35.
One of the prototypes has been making runs across the country. Yes, it’s just a prototype for now. But it’s a roadworthy vehicle.
More importantly, it’s trivial to do the napkin calculations to see that a 500 mile range is reasonable. It’s not hard to show that you need around a 2 MWh battery, and that this fits into the mass and size constraints of a semi. If they can’t pull it off it’s because they fucked up the design somewhere, not because there’s anything that prohibits it even with current technology (though to meet their price targets, I suspect they’re depending on cell prices going down).
Same goes for raw performance. EVs have very predictable performance because the drivetrain is so simple. Although an unloaded 5 second 0-60 isn’t important for a semi, a 20 s 0-60 for a fully loaded one does make a difference, as is the ability to sustain its speed up a grade.
Sorry, that should be a 1 MWh battery. Efficiency should be around 2 kWh/mi, giving a 500 mi range or thereabouts. It’s just 10 Model S batteries. Not really a big deal.
Only 3.5 M class 8 trucks (i.e., a typical “semi”). Still a lot, though.
BTW, I don’t necessarily mean that semis will even have the highest adoption rate, but rather that they’ll be one of the first cases where otherwise perfectly-good trucks are replaced because they’ve become uneconomical to operate. Most cars don’t drive enough miles to justify early replacement just to save on fuel costs, but for trucks the tradeoffs are different.
You left out team drivers in your equation. You also left out parking remotely which requires a generator. Then there are recharging stations at every truck stop. Drivers can’t be hunting for a place to charge. They need to deliver as soon as possible. And finally, your scenario lacks a power grid that can handle the increase load as well as a tax base to keep the roads in good repair.
A great many things need to change to make this work.
I expect deep learning to make much more profound changes than this.
Current control algorithms look something like “semi-blindly make movement 1, 2, 3. CV checkpoint. Make movement 4. CV checkpoint. If any unexpected events, stop”.
The machine doesn’t have an internal environment model of the steps it is doing. It can’t figure out how to clear it’s own faults. Generally if the raw materials or lighting change it won’t be able to just learn how they look now, the computer vision will just report a fault and the line will stop.
An algorithm that is eventually feasible, using deep learning as an enabling component, would be more like :
a. Practice with robotic manipulators and raw materials of similar classes to the type used for the next job. (in reality the robotic system would purchase and download this practice data from some online marketplace, it wouldn’t waste time re-creating it)
b. From practice, develop a predictive model of the likely state transitions for each action taken. State transitions are things like “if I take action <hit this ball>, the new state of the ball is <this velocity vector>”.
c. With this predictive model - the model itself uses deep learning neural networks and is based on real data, so it will be very accurate - plan possible assemblies that will result in the final product given in a set of manufacturing schematics.
So a robot inserting a car seat will not need to be told how to do it, only what the goal is and some metrics for scoring different approaches.
d. Try a variety of strategies in the real world to accomplish the task.
e. Go back to (b) until incremental improvements become negligible.
This will be a profound change. It will ultimately allow automating the manufacture of basically everything, and new products, even prototypes, will be made by robots, not people.
In addition I think that the infrastructure to support all this will be a common pool of industry standard robotic manipulators and sensors. (anyone selling a robot that doesn’t meet industry standards will get crushed out of the marketplace, same fate as all the non Android/iOS phone ecosystems). A common pool of many software algorithms and datasets you’ll be able to apply to a given problem.
And eventually, this may be the seeds of a “real” AI. In this ‘marketplace’ for robot control, people might start selling ‘meta’ robotic planning apps. Software that can self-optimize a design. That can choose a correct neural architecture for a given dataset. Eventually someone else might build on those “meta” algorithms with a more meta algorithm than that, and bolt on algorithms that can communicate with humans in natural language or internalize a model for carrying on a conversation, and we’ll start to meet semi-credible AIs.