I think we'll know fairly quickly whether the FSD branch is going to be a distruptive change. Either the NN will learn quickly - which means the number of interventions per driven mile dropping rapidly - or it just continues the slow pace of development we've seen with AP and we all go back to not expecting autonomy in the next 10 years..
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD / AP Rewrite - turning the corner?
- Thread starter MrBadger
- Start date
Glan gluaisne
Active Member
Perhaps worth remembering that the key driver behind the development of autonomous vehicles isn't personal use cars at all, it's the ride-hailing market. Last time I looked this was estimated to be worth over $3 trillion, so a very big and attractive target.
Ride-hailing has the potential to be a massive disruptor to public transport, given that we already have the communications and geo-location/tracking infrastructure needed to enable it. Uber was one of the first to see that combining this existing infrastructure with current mini-cab services could be very profitable. Add in autonomous vehicles, so that any ride-hailing company doesn't have to deal with tens of thousands of drivers, and whoever gets in first will make a great deal of money.
It's no secret that this is the business that Elon Musk is aiming at. The production of cars is just an interim step for Tesla, I would guess that, within ten years, Tesla's main business will be producing what we'd call taxis today. Selling cars as an interim measure has several really big benefits for Tesla. It generates cash, both from sales and investment (the latter the most important, I suspect). It provides a mountain of mapping and driver/vehicle/other road user behavioural data (Tesla probably have more real-world vehicle mapped miles than anyone else, including MobilEye). The cars also allow Tesla to experiment with a large fleet of vehicles, by changing the software every few weeks and measuring the effect.
Ride-hailing has the potential to be a massive disruptor to public transport, given that we already have the communications and geo-location/tracking infrastructure needed to enable it. Uber was one of the first to see that combining this existing infrastructure with current mini-cab services could be very profitable. Add in autonomous vehicles, so that any ride-hailing company doesn't have to deal with tens of thousands of drivers, and whoever gets in first will make a great deal of money.
It's no secret that this is the business that Elon Musk is aiming at. The production of cars is just an interim step for Tesla, I would guess that, within ten years, Tesla's main business will be producing what we'd call taxis today. Selling cars as an interim measure has several really big benefits for Tesla. It generates cash, both from sales and investment (the latter the most important, I suspect). It provides a mountain of mapping and driver/vehicle/other road user behavioural data (Tesla probably have more real-world vehicle mapped miles than anyone else, including MobilEye). The cars also allow Tesla to experiment with a large fleet of vehicles, by changing the software every few weeks and measuring the effect.
phil4
Member
You're right about Uber and Tesla both saying their aiming for automated taxi services.
What perhaps we need to keep in mind is while it can reduce the pollution with regard to creating cars, it almost certainly increases any use based pollution. My car goes with me two and from work and sits idle waiting for me while I work. If I use a taxi it has to travel to me to use, travel away from me for the next person, then back to me to pick me up, and then away from me again to it's next job. I'm not saying it'll increase massively on average, but the more rural/remote you are, the more likely it'll go a lot lot further.
The other thing to consider is the cost. I realise that drivers are a large part of the cost, but equally if the above inefficiency equates to higher costs, both Uber and Tesla won't get too much extra business outside of cities. Or put another way, I don't use Uber/A taxi for my daily commute because a) it costs too much b) takes too long to get to me. This can be solved by automation, but it'll not be easy, especially the last one.
What perhaps we need to keep in mind is while it can reduce the pollution with regard to creating cars, it almost certainly increases any use based pollution. My car goes with me two and from work and sits idle waiting for me while I work. If I use a taxi it has to travel to me to use, travel away from me for the next person, then back to me to pick me up, and then away from me again to it's next job. I'm not saying it'll increase massively on average, but the more rural/remote you are, the more likely it'll go a lot lot further.
The other thing to consider is the cost. I realise that drivers are a large part of the cost, but equally if the above inefficiency equates to higher costs, both Uber and Tesla won't get too much extra business outside of cities. Or put another way, I don't use Uber/A taxi for my daily commute because a) it costs too much b) takes too long to get to me. This can be solved by automation, but it'll not be easy, especially the last one.
Glan gluaisne
Active Member
I did some rough sums a while ago, regarding the relative cost of taxis versus car ownership. We live in a rural area, a bit over 10 miles to the nearest sizeable town. For us, using taxis would have been cheaper than owning cars, by a fair bit. That's at the relatively high cost of taxis at the moment. I've no doubt that autonomous taxis will be cheaper. The reason we don't use taxis is simply because we both like driving. Take away that, by making cars fully autonomous and we may as well use taxis instead.
phil4
Member
I do get that, and I think it'll depend on who and what you drive for. You're right that if cost is solved, more people will. But for me, I don't want to be waiting for a taxi now or when it's autonomous, even if it costs less. A few minutes, no probs. 20-30, I'd just drive.
I think massive savings may well of course drive changes in attitude (ie. call the taxi 30 minutes early). But equally I'm aware that Tesla/Uber want to make money, so cost can't go down too much. It'll be interesting to see how it pans out. I'm defo not against it, in fact I think the idea of automation is brilliant I want a piece
It still increases pollution though
I think massive savings may well of course drive changes in attitude (ie. call the taxi 30 minutes early). But equally I'm aware that Tesla/Uber want to make money, so cost can't go down too much. It'll be interesting to see how it pans out. I'm defo not against it, in fact I think the idea of automation is brilliant I want a piece
It still increases pollution though
Glan gluaisne
Active Member
If had to try and predict what might happen, then I'd say that it will be city areas where autonomous vehicles will initially have far and away the greatest impact. I think it's entirely possible that we may have virtually no private cars, and no driver-operated buses and taxis in some of our cities within the next ten to fifteen years, maybe less if legislation continues to make driving in cities more expensive, year on year. There will be a lot of resistance, from taxi and bus drivers, but it seems inevitable that this will happen. Having intelligently controlled autonomous vehicles, that can provide safe, quick and reliable transport for individuals or groups of people would become popular in just the same way that Uber has. Removing drivers from vehicles in cities removes a great deal of the problems, like thoughtless parking, the congestion that causes, the need for large car parks full of vehicles that aren't used for >90% of the time, etc, as well as solving the vehicle pollution problem, if the replacement system is electric.
When it comes to transport outwith cities and large towns, things get a great deal harder to predict. I suspect that people like us, who live ~10 miles from the nearest reasonably sized town, might be the last to move away from car ownership. It's quite possible that the personal car market shrinks to the point where owning a car gets significantly more expensive than the alternatives, especially as those alternatives reduce in relative cost, due to their popularity in urban areas.
There are plenty of times right now when, if there was a convenient and affordable alternative to using a personally owned car I would use it. For example, our local pub is about 4 - 5 miles away, so we drive there, which then means one of us not drinking. Going to the theatre (when this plague is under control) is a >20 mile round trip, and in winter, or bad weather, I'd much rather not drive if there was a convenient alternative. Things are already changing, over the past few months we've completely stopped going to shops, and do almost all our shopping online, with all of it, except the local farm shop order, being delivered to the door.
When people talk about disruptive technology, like autonomous transport systems, I think many may not realise just how disruptive it may be. It's entirely possible, perhaps even very likely, that personal car ownership will just cease to exist. The primary things that make car ownership attractive now are the convenience (perhaps the most significant factor) and the effectiveness. The cost of car ownership is already high, higher than the cost of most other forms of transport, in terms of total cost per mile (not just fuel cost). Take a Model 3 as an example. Over 5 years it will probably cost ~£20k to £30k in depreciation, ~£2k to £3k in insurance, and for 10,000 miles/year around £2k in electricity, so around £3/mile. If something else can be every bit as convenient and effective, at significantly less than about £3/mile, then pretty soon most people would just stop owning cars.
When it comes to transport outwith cities and large towns, things get a great deal harder to predict. I suspect that people like us, who live ~10 miles from the nearest reasonably sized town, might be the last to move away from car ownership. It's quite possible that the personal car market shrinks to the point where owning a car gets significantly more expensive than the alternatives, especially as those alternatives reduce in relative cost, due to their popularity in urban areas.
There are plenty of times right now when, if there was a convenient and affordable alternative to using a personally owned car I would use it. For example, our local pub is about 4 - 5 miles away, so we drive there, which then means one of us not drinking. Going to the theatre (when this plague is under control) is a >20 mile round trip, and in winter, or bad weather, I'd much rather not drive if there was a convenient alternative. Things are already changing, over the past few months we've completely stopped going to shops, and do almost all our shopping online, with all of it, except the local farm shop order, being delivered to the door.
When people talk about disruptive technology, like autonomous transport systems, I think many may not realise just how disruptive it may be. It's entirely possible, perhaps even very likely, that personal car ownership will just cease to exist. The primary things that make car ownership attractive now are the convenience (perhaps the most significant factor) and the effectiveness. The cost of car ownership is already high, higher than the cost of most other forms of transport, in terms of total cost per mile (not just fuel cost). Take a Model 3 as an example. Over 5 years it will probably cost ~£20k to £30k in depreciation, ~£2k to £3k in insurance, and for 10,000 miles/year around £2k in electricity, so around £3/mile. If something else can be every bit as convenient and effective, at significantly less than about £3/mile, then pretty soon most people would just stop owning cars.
Steven Peeters has some good videos on the UN ECE 79 regs and their implications. All his videos are normally excellent although not if you only have a spare 10 minutes (not a criticism, just acknowledgement of his depth of detail)
Starting around 10:30 and an excellent worked example just after 13:00
And an update on regs at around 0:45 seconds
JupiterMan
Active Member
There's some in depth reading on the economics of autonomy in the RethinkX report, which considers the Transport as a service implementation from 2020 to 2030:
https://static1.squarespace.com/sta...fba6/1509063126843/RethinkX+Report_102517.pdf
A long read and 3 years old now, but still a big eye opener... the report looks into the economics and wider societal/geopolitical changes of the "transport as a service" model and they are compelling.
This report was one of the reasons I bought TSLA stock back in the day (and why I will continue to hold for a few years yet).
People currently cling to the ownership model and it will be interesting to see how this changes as we go forward and autonomy matures...
Glan gluaisne
Active Member
IIRC, Tesla weren't that keen on people owning the cars they make outright at one time. I can't recall where I heard it, but I seem to remember Elon Musk suggesting that people would only be able to rent Tesla's cars at one time, years ago, not buy them outright. I'm pretty sure this was one of the things that fell by the wayside when FSD took longer than originally planned, but it struck me that the reasoning behind it was so that Tesla would already own the world's largest autonomous vehicle fleet when Level 5 was finally cracked. This would then put Tesla in a very good position to just shift into the personal transport as a service model, ahead of every other auto manufacturer.
Whether Tesla will be the first to achieve routine Level 5 operation on public roads, or whether it be one of the other half dozen or so companies working on it, I wouldn't like to guess. Tesla aren't at the head of the pack right now, there are other developers who have significantly more capable systems already working well. The question really comes down to whether Tesla's gamble on doing the majority of the sensing using cameras pays off. A couple of years ago, when cameras were far and away the cheapest sensor option, I'd have said that Tesla had definitely made the right decision. Things aren't so clear now, though.
Over the past two years the cost of LIDAR, for example, has reduced by over an order of magnitude - Waymo's sensor suite, for example, was estimated to cost ~$7,500 last year, whereas it cost over $80k a couple of years earlier. Prices are still dropping, such that a LIDAR with similar resolution and range to a camera is only about three times the cost, and there's every chance that LIDAR may well reach cost parity with cameras within the next year or so. I'm not yet convinced that LIDAR is the best solution for vehicles, though, as it's affected by the environment in much the same way cameras are.
What we don't know is what is going on behind the doors of some of the less publicity-oriented companies. I know that there are some RADAR-like technologies, that have been used for portable, high resolution, 3D sensing in military applications, and some of those, if they could break out of the constraints imposed by the funding sources for those developments, might just transform the way autonomous vehicles sense the world around them.
The whole industry has been on this trajectory for ages. Its also the reason why many use the term mobility rather than transport. Far more wide reaching than just getting from A to B. Its a whole rethink of if you need to get from A to B in the first place or just need access to B. It just a small piece of the smart city jigsaw . Where I use to work was heavily involved in this. But its going to need some big mindset changes especially if things like mobility as a service is going to start to be more widely accepted.
Anyway, this is well OT.
Anyway, this is well OT.
phil4
Member
What puzzles me immensely on this subject is trains.
When you consider how hard FSD is, imagine doing it with trains... should be an absolute doddle. But apart from the odd monorail, or the Heathrow T5 pods... none of the main train lines are even talking about it. And that's despite some new rolling stock being rolled out over the past few years (GWR) etc.
When you consider how hard FSD is, imagine doing it with trains... should be an absolute doddle. But apart from the odd monorail, or the Heathrow T5 pods... none of the main train lines are even talking about it. And that's despite some new rolling stock being rolled out over the past few years (GWR) etc.
The stopping distance of a train is much longer than any feasible radar can see and react to which is why I think there are no FSD type trains currently.
ERTMS is probably the closest solution however it's not driverless, more it uses the positioning data of the train and others around it to work out which part of the track it can safely be on and at what speed.
https://www.ertms.net/?page_id=23#:...s a European,capability of legacy ATP systems.
The level 2 standard of ETCS has been deployed on a few small areas of the UK network, the East Coast Mainline will be the largest deployment, work starts next year.
phil4
Member
I do get that, the stopping distance is also further than a driver can see... perhaps the driver cost is minimal, so little reward overall, but genuinely seemed like a much much easier problem to solve.
Glan gluaisne
Active Member
DLR had the big advantage of being designed as a fully autonomous, driver-less system, so there was no need to try and deal with the plethora of complex signalling and control systems used by conventional trains. I strongly suspect you're spot on about the union and rail industrial issues, too. A friend of mine is a driver for SWR (was SWT) and I avoid the topic of unions whenever we chat. He earns more than many airline captains, believe it or not.
It's absolutely the unions. They can't even agree that drivers should open and close the doors as it puts the role of the guard at risk. The technology is there and has been for a long time, the Victoria Line runs semi-autonomously with human 'Train Operators' opening/closing the doors and has done since 1968.
So far, I have been very encouraged by watching many of the videos showing the new 4D FSD on city streets. I now expect that the cars will be able to cope with roundabouts and even slow down for speed bumps and avoid potholes, as well as drive down roads with no lane markings. It definitely still needs further improvement but most people should surely agree the new release demonstrates a genuinely large advance in functionality over what we have seen so far.
The big question remains how long it will take to improve reliability from here such that it 'just works' and you can feel more relaxed about letting the car drive you, as you would if another experienced human driver offered you a lift. Right now it still feels more like my first experiences in the car with my children driving, right after they got their license...
I think one of the most interesting things I have read about the Tesla system however is not the status of their FSD system today, impressive as the new 4D rewrite appears to be, but rather the 'dojo' machine learning system they have been quietly developing. My understanding is that this dojo system represents even more of a 'quantum leap' in technology than its product ie FSD. It will automate the process of supervised training of the neural net that will iron out the edge cases and ultimately deliver an autonomous driving system that exceeds the performance of human drivers on city streets. I wish I could find out more about this, as I suspect it could end up being key to whether the Tesla FSD system evolves from its current state to reliable autonomy really fast, or whether the process will take much longer and require further hardware upgrades along the way.
JupiterMan
Active Member
You're right, DoJo will be the next big thing for Tesla FSD development. Not much is known from what I can find other than some throw away headlines in various speeches here and there. Would love to learn more too.
It is interesting to me that the FSD Beta testers are reporting significant improvements already since the release only one week ago, which I think shows the new re-write should allow much quicker development than what we experienced before.
With Dojo online that is likely to become exponential. I have a feeling we will be at self driving a lot quicker than people think... it is just the governments and regulators who will need to catch up....
Wol747
Active Member
I accept that this is one objective, but I believe Musk's oft repeated concept - the three stage production of EVs to jump-start the widespread adoption of renewable energy for transport.
I wouldn't have spent over $160k on a car if I wasn't convinced of the rightness of that: just my small contribution to the objective.
Similar threads
