And this is the beginning of the end of bus driving as a profession. Sure, this is in well defined, short (and private) routes. First Santa Clara, then Stanford shuttles, then airport shuttles, then Googlebus drivers.
Enjoy the raises while you can.
They are developing platforms that will operate inside contained environments. What you are referring to is very different from what they are doing. An autonomous vehicle that operates on a public road faces much more challenges than one that does so in a contained environment. While working on my own autonomous car (yes, I bought a car for the sole reason to try and do this), I realized that the simplest route would be to do as they did. It can operate in a closed loop, similar to line following robots. Eerily similar to automated trains in airports. Its a challenge to build a safe autonomous vehicle, a bit less so in the way they are doing it. Not trying to diminish their efforts though. I have first hand experience of how hard this simpler approach is.
One obvious question, why aren't more subways fully automated? It's not like they can't implement remote control override for emergency situations... Is it still a psychological issue (with riders)?
Because most subways run on legacy infrastructure that would be difficult to upgrade/replace. RF repeaters for emergency remote control for example would need to be installed, certified and thoroughly tested.
Given that drivers aren't a big cost centre for subway operations in comparison to maintenance etc. the cost/benefit doesn't make sense.
I think the pensions are a sizable part of their expenses. If they automate their retired workforce age out and there is no endless retirement pool to contribute to. So this would be a significant savings.
They will be. I was thinking of this earlier today, and there is still some coordination between schedules, doors closing with people still trying to get on, etc. etc. In Sydney, you still have people blowing whistles to signal to the driver that it is safe to close the door and go.
I'm wondering if that is maybe a more difficult problem to solve than driving. In driving, you know the direction cars 'should' be going, you need to account for lane changes and turns, etc. but for the most part they are expected. I think in an open environment people are much less predictable in their movements.
In metro systems, 50 percent of service affecting failures come from passenger exchange. Majority of the remaining failure come from vehicle failures. Failures due to driving mistakes are very few.
Passenger flow management is the hard part of automated mass transit for performance. It has to be safe yet efficient. It's indeed surprisingly difficult.
Automatic new line are easy but it's harder to automatize old line without interrupting traffic. It has been done for line 1 in Paris between 2007 and 2011 (https://en.wikipedia.org/wiki/Paris_M%C3%A9tro_Line_1) but all other old line in Paris have still not been automatized.
The first fully automated subway line was in New York, in 1959. The Times Square - Grand Central shuttle was fully automated for six years.[1] But, per union rules, they still had someone on board. A major platform fire at the 42nd St station damaged the rolling stock, and when the station was rebuilt, the one-off automated system was not rebuilt.
The main reason for trying it on that line is that the trip is very short, and the train reverses at each end. The train operator's walk though the train to get to the other end caused delays. Paying for two drivers, one at each end, was a bit much.
No doubt it is hard, our previous experience helped us a lot, we did make a fresh prototype vehicle in USA in just 2.5 months including procurement. Working with newer sensor and writing different perception algrithmso for current use case and even we used a different motion planning approach. We never got stuck anywhere, the only problem is time vs skilled manpower. We just wish we could have more people in our team to build more features sooner and ironing out more corner cases which have not prioritized due to lack of time. There are so many module and so many things we want to improve, and we know how, just need few more hands.
We our-self feel very proud to have been able to come to USA, hire interns, get the basics of every module right enough and integrate on time, including getting drive by wire and other retrofitting done and tested and stated piloting completely autonomous in unimaginable short time frame, thanks to the push from YC partners.
I'm a student at SCU and I'm excited to see these on campus; it's not a big enough campus to merit having it's own shuttle system, but it should be a perfect testbed for the platform.
Your comment echoes a doubt I have. Is there really a sizeable market for this? It's not such a new concept as one may believe. For example, in some offices, such robots carried around the mail 15 years ago already.
We will meet soon then :) .. hope you are exited to get autonomous rides. University authorities are excited to deploy such technology both to stay ahead and have cool inspiring tech in campus and also to help elderly people like some professors and visitors (who regularly come to church), or anyone who has trouble walking and they can even use it for logistics within campus. When something becomes much cheaper in the industry use-case possibilities explode. So lots of reasons even for a small campus and everyone is excited about it.
We do need real funding. The more we can make the sooner and better we can make things. Having lot of modules and test sites we can easily parallelize work of both testing at different sites and collecting more data and working on improving more modules at the same time. So basically we can truly utilize higher amount of funding
There will be an autonomous shuttle starting (testing) october 2015 in Wageningen, the Netherlands. Production use expected May 2016. Big caveats though: will drive only 25 km/h, won't operate at night or in bad weather. It'll shuttle between Ede trainstation and the WUR campus.
Fortunately we are starting at Bay area where bad weather is less frequent and we can also operate on mild bad weather and have done lot of testing at night (in fact that was our prime hours of testing). We have also worked in bright sunny daylight in scorching heats and overcome issues with sensor noise in both conditions, only hoping to make even smarter algorithms and deals with more wider range of situations and we continue our pilots and paid deployments.
Good luck! I know how hard a problem this is, as I was part of the team that worked on the driverless taxis now operating at Heathrow airport[1] (and which have safely carried over a million people since launch).
You might also find the company's library of research papers a useful reference [2].
Thanks for the reference. We did talk to one of the first employees of ultraglobal once and learnt its limitations and the challenges it faced. Autonomous vehicle technology is definitely complex but deployable for low speed vehicles in controlled environments. The recent developments in the autonomous Robotics space and a significant reduction in sensor prices make this the best time for such a product.
Also the fact that ultrapods carried millions of passengers despite it's environment being constrained to rails on either side, proves the point that completely autonomous vehicles that can move freely around the campuses will have a greater use case.
This is cool. Yet the primary efficiency over conventional shuttles appears to be limited to lower labor costs. Unlike the taxi market, it does not appear to reduce vehicle count or improve passenger experience via better service.
Better service in a regular route context comes mostly from dedicated right of way and platforms. The classic example is Curitiba's bus system in lieu of a subway system It could not afford. The real change for campus shuttle service will be if automation leverages itself into dedicated shuttle rights of way as a safety measure or direct commitment to improved service.
26 comments
[ 3.1 ms ] story [ 66.3 ms ] threadOne obvious question, why aren't more subways fully automated? It's not like they can't implement remote control override for emergency situations... Is it still a psychological issue (with riders)?
Given that drivers aren't a big cost centre for subway operations in comparison to maintenance etc. the cost/benefit doesn't make sense.
I'm wondering if that is maybe a more difficult problem to solve than driving. In driving, you know the direction cars 'should' be going, you need to account for lane changes and turns, etc. but for the most part they are expected. I think in an open environment people are much less predictable in their movements.
Thoughts?
Passenger flow management is the hard part of automated mass transit for performance. It has to be safe yet efficient. It's indeed surprisingly difficult.
The main reason for trying it on that line is that the trip is very short, and the train reverses at each end. The train operator's walk though the train to get to the other end caused delays. Paying for two drivers, one at each end, was a bit much.
[1] http://www.nycsubway.org/wiki/IRT_Times_Square-Grand_Central...
Yes, they are less flexible because they need a fixed track, but this is just a step that loosens the "well defined" part.
You might also find the company's library of research papers a useful reference [2].
Looking forward to seeing your progress!
[1] http://www.ultraglobalprt.com/
[2] http://www.ultraglobalprt.com/about-us/library/papers/
Better service in a regular route context comes mostly from dedicated right of way and platforms. The classic example is Curitiba's bus system in lieu of a subway system It could not afford. The real change for campus shuttle service will be if automation leverages itself into dedicated shuttle rights of way as a safety measure or direct commitment to improved service.