2014 MATC Lecture Series - Dr. Chris Schwarz, "Towards Autonomous Vehicles"

MATC Author

11/10/2017 Added

7 Plays

Description

Searchable Transcript

Search:

[00:00:00.394]Good morning, this is the first seminar
[00:00:03.473]in MATC Spring 2014 Webinars three series.
[00:00:07.003]We are really excited to have Dr. Chris Schwarz from
[00:00:14.188]NADS to be presenting on Towards Autonomous Vehicles.
[00:00:18.194]Dr. Schwarz received his B.S. degree
[00:00:19.934]in electrical engineering
[00:00:21.850]from the University of Illinois in 1990,
[00:00:25.301]he has received a Ph.D. degree in electrical engineering
[00:00:28.276]from the University of Iowa in 1998,
[00:00:30.992]and he has worked as a Research Engineer
[00:00:33.626]at National Advanced Driving Simulator since 1998.
[00:00:39.635]This seminar is also
[00:00:46.442]related to a recent MATC project
[00:00:48.222]which Dr. Schwarz was working on,
[00:00:50.207]and the final report was submitted in 2013,
[00:00:53.387]and I'd like to hand it over to Dr. Schwarz to
[00:00:57.380]start his presentation, thank you.
[00:01:01.505]Hi, good morning.
[00:01:05.556]And just another note on the audio,
[00:01:07.744]you were kinda soft there, so maybe you just need to
[00:01:10.646]speak up when you're doing the questions later,
[00:01:15.168]that would be great.
[00:01:18.383]So yeah, thank you all for coming,
[00:01:21.593]I was very excited to do this project,
[00:01:25.504]and I'm very excited to share the results from it.
[00:01:29.927]The title of the report and of the presentation
[00:01:32.927]is Towards Autonomous Vehicles,
[00:01:38.296]and I just wanna start with some acknowledgements.
[00:01:43.948]As was said, this was a project
[00:01:46.351]for the Mid-America Transportation Center,
[00:01:49.932]it was a one year project to survey the field,
[00:01:53.164]do a literature survey, and write a report.
[00:01:56.263]My Co-PI was Professor Geb Thomas
[00:01:58.474]in the Department of Mechanical Industrial Engineering,
[00:02:02.895]and over the course of the year
[00:02:05.160]we had four undergraduate students
[00:02:08.457]who were able to help me out doing,
[00:02:12.090]reading papers and developing some scenarios,
[00:02:16.112]and they were all really great.
[00:02:19.629]So the work products, there are a couple that
[00:02:22.679]you can get to, the first is the final report,
[00:02:26.669]it's on the MATC website at UNL,
[00:02:34.267]and the second is the body of research, the papers,
[00:02:40.842]are all collected in a Zotero group,
[00:02:43.941]and that's a public group, you can all go and
[00:02:49.093]actually grab any or all of the papers
[00:02:52.151]that we collected, and that's available at zotero.org.
[00:02:59.813]I will refer you to those for all the references
[00:03:03.719]that were used, 'cause I don't include
[00:03:06.139]a lot of the references in this slide.
[00:03:09.801]So autonomous vehicles are all the hype right now,
[00:03:15.193]you may not see them on the road a lot, or at all,
[00:03:19.146]but it's hard to not see reports
[00:03:22.651]and news stories about them, magazine articles,
[00:03:27.097]so there's a lot of excitement, and it seems
[00:03:31.822]somewhat inevitable that they'll be coming our way
[00:03:34.346]at some point in the future.
[00:03:37.893]What I'd like to do with this talk
[00:03:39.686]is cover a little bit of the history,
[00:03:42.653]a little bit of the current state of the art,
[00:03:46.093]and the different elements of autonomous vehicles,
[00:03:50.637]then talk about some of the challenges that they face,
[00:03:54.874]and a little bit of discussion on a couple of issues
[00:04:02.237]that I pulled from our research,
[00:04:05.187]and then we'll have some time for Q&A.
[00:04:10.069]The terminology, automation, autonomous,
[00:04:13.205]automated, semi-autonomous, there are a lot of terms,
[00:04:16.880]the government now prefers automated vehicles
[00:04:20.049]because it captures all of the different levels
[00:04:24.848]in between fully autonomous and fully manual.
[00:04:30.515]I'm not gonna be strict in my use of the terminology,
[00:04:33.270]I'll use autonomous, I'll use automated sometimes.
[00:04:39.623]And of course it's impossible for me
[00:04:44.710]to present at a level that will make everyone happy,
[00:04:50.433]I feel that this is an intermediate presentation,
[00:04:54.809]not really for complete novices,
[00:04:58.861]but those of you who may be very familiar with the field
[00:05:04.359]might think it's a little too simplistic or a summary,
[00:05:09.938]so sorry if I don't hit everyone's sweet spot,
[00:05:12.545]but I hope it'll be interesting nevertheless.
[00:05:15.774]So what's the motivation for autonomous vehicles?
[00:05:22.283]There are a lot of visions for the field,
[00:05:26.107]one of the primary ones is safety.
[00:05:28.768]There's still over 30,000 people who are killed every year,
[00:05:32.752]and over 90% of those are due to driver error,
[00:05:37.001]and beyond fatalities there are many more injuries
[00:05:39.498]and lots of property damage as well.
[00:05:41.785]So the vision for autonomous vehicles
[00:05:44.411]is a crashless society.
[00:05:48.931]If you can eliminate all of the crashes
[00:05:52.477]due to driver error, then that's gonna be a vast improvement
[00:05:58.220]and save lives.
[00:06:00.597]So the second area would be the area of mobility.
[00:06:04.732]One of the things that autonomous vehicles
[00:06:06.792]can provide for us is the ability to
[00:06:09.818]put a lot more cars on existing roads,
[00:06:14.667]then we would not have to keep adding to the infrastructure,
[00:06:18.884]adding more lanes, adding more roads.
[00:06:24.317]So this'll require some fairly high market penetration
[00:06:28.245]for the technology, 'cause you wouldn't wanna necessarily
[00:06:31.466]be driving manually on a road
[00:06:34.613]that's so densely packed with vehicles,
[00:06:39.148]but the other great thing that it will hopefully do
[00:06:42.196]is provide greater access for those who can't drive,
[00:06:45.856]the elderly, disabled, children.
[00:06:51.221]It would provide a lot of freedom
[00:06:52.802]to a lot of people who don't have it.
[00:06:55.093]And then the third area is sustainability.
[00:06:59.346]There are some examples such as platooning
[00:07:02.779]where we've been able to estimate
[00:07:04.558]the potential fuel savings, maybe 20%.
[00:07:08.437]There are other ways though that it will contribute
[00:07:12.708]to a sustainable economy or society,
[00:07:17.266]by eliminating traffic jams, reducing trip times,
[00:07:21.303]potentially in the future reducing ownership of vehicles,
[00:07:24.540]reducing the needs for parking garages
[00:07:26.664]and parking spaces, and being able to convert
[00:07:29.592]some of those to other uses, so it's very exciting,
[00:07:33.131]the potential is very exciting.
[00:07:37.206]As I talk through the historical part of this
[00:07:42.246]I wanna emphasize the different cycles of innovation
[00:07:47.885]that autonomous vehicles have gone through,
[00:07:50.221]but really that every technology must go through.
[00:07:56.243]It starts out with very exploratory and basic research,
[00:07:59.706]and eventually works its way into commercial products,
[00:08:04.053]but it doesn't just happen once,
[00:08:05.600]it then repeats over and over again,
[00:08:08.143]but every time it goes through that cycle
[00:08:12.693]it gets a little more advanced
[00:08:13.772]and a little bit more polished,
[00:08:15.493]and that's what we've seen over the last 40 or 50 years.
[00:08:22.737]I wanna introduce this little graphic
[00:08:25.694]that you'll see on the next few slides that indicates
[00:08:32.866]where the impetus and the funding
[00:08:36.345]for a particular project or program came from,
[00:08:42.510]and I'm including academic-inspired projects
[00:08:46.042]and non-military government, private industry,
[00:08:49.664]and then military.
[00:08:54.516]And that's what it will look like on the slides.
[00:08:57.371]So one of the earliest was a project that was done
[00:09:02.842]in the 50s in Nebraska,
[00:09:07.505]a collaboration between the state of Nebraska and RCA,
[00:09:12.901]and I just love the quote,
[00:09:14.844]"Driving will one day be foolproof, and accidents unknown."
[00:09:18.451]So it took a while for it to get there,
[00:09:21.409]and we're still not there,
[00:09:23.131]but the promise certainly seems real today,
[00:09:26.587]and it probably seemed very science fiction back then,
[00:09:31.470]but that was a very future-looking experiment back then.
[00:09:40.661]Really in the 80s and 90s, and going up 'til today,
[00:09:45.241]there are examples like Carnegie-Mellon
[00:09:48.229]with their NAVLAB, which now has, it's in its 11th version
[00:09:55.821]for testing and developing autonomous vehicle technology.
[00:09:59.382]And NAVLAB, you could think of it as a platform
[00:10:02.529]upon which they've done several projects
[00:10:07.374]that fit onto that platform like RALPH,
[00:10:11.694]which in the mid-90s drove over 3,000 miles,
[00:10:17.857]and one trip from Pittsburgh to DC,
[00:10:21.857]96% of that was steered autonomously.
[00:10:26.068]So the problem of following a road and following a lane
[00:10:30.082]has really been...
[00:10:36.655]It was done a long time ago,
[00:10:38.744]and a lot of the effort since then has been
[00:10:43.542]dealing with edge cases, improving with liability,
[00:10:47.972]but this is an example that sometimes the challenge
[00:10:51.028]isn't the basic technology, it's other areas.
[00:10:55.797]So after that came along a government initiative
[00:11:01.565]with automated highway,
[00:11:05.112]and this National Automated Highway System
[00:11:07.821]was a program that the DoT started, also in the mid-90s,
[00:11:12.975]and it went strong for about three years,
[00:11:17.337]and culminated in a demonstration in California
[00:11:22.755]of some autonomous driving,
[00:11:24.883]and one of the features that they emphasized
[00:11:29.999]was platooning, so this is an example of platooning.
[00:11:34.459]And with the benefits of fuel efficiency,
[00:11:37.160]of safety, of convenience,
[00:11:40.065]you can take your hands off the steering wheel,
[00:11:43.121]take your feet off the pedals, and just ride along.
[00:11:49.249]The difficulty with this, that people realized,
[00:11:56.214]was that it was gonna require enormous investments
[00:11:58.568]and infrastructure to deploy something like this,
[00:12:02.039]so really what happened was after that,
[00:12:06.182]it changed focus a little bit and started to emphasize
[00:12:11.389]individual systems.
[00:12:17.282]For example, collision warning systems,
[00:12:19.395]lane departure systems, and this is really where
[00:12:24.292]a lot of these systems got their start,
[00:12:28.817]and we see them today in cars on the road,
[00:12:33.724]so this turned out to be a really good incubator
[00:12:38.192]for a lot of these technologies.
[00:12:43.262]Now, a little bit later in the mid-2000s,
[00:12:47.911]we started to have well-known DARPA Grand Challenge
[00:12:53.913]that seemed to capture the imagination of a lot of us,
[00:12:57.067]it did me for sure.
[00:13:01.048]The first time nobody won,
[00:13:03.153]in fact I don't think anybody got over a few hundred feet,
[00:13:07.416]but they ultimately did have a winner in 2005,
[00:13:10.352]and then in the more difficult Urban Challenge
[00:13:14.664]the Carnegie-Mellon vehicle won.
[00:13:20.108]So going back to now government initiative,
[00:13:24.009]in the mid-2000s the government started looking at
[00:13:28.552]connected vehicles, or back then it was called
[00:13:31.796]Vehicle Infrastructure Integration,
[00:13:35.818]and it was around that time
[00:13:37.900]that the government had some frequency band
[00:13:42.300]reserved for them for the DSRC networks,
[00:13:45.612]and that's been ongoing for 10 years,
[00:13:48.892]and recently NHTSA announced its intention
[00:13:54.335]to enable this technology,
[00:13:57.791]and at some future date, to begin to mandate
[00:14:01.692]and regulate it in light cars.
[00:14:04.809]So this is pretty exciting,
[00:14:08.808]there are several safety applications
[00:14:11.600]that have been identified that this could really help with,
[00:14:16.022]like forward collision, intersection movement assist.
[00:14:21.086]The great thing is that it's not limited to line of sight,
[00:14:23.885]but it can receive a message from a car
[00:14:26.881]that's two cars ahead,
[00:14:28.737]or maybe it's geared by something.
[00:14:34.877]So this provides some benefits that I hope we'll see,
[00:14:42.499]and there's been a safety pilot going on up in Ann Arbor.
[00:14:46.475]Connected vehicles, people are also interested
[00:14:51.169]in the benefits of mobility and sustainability
[00:14:54.097]with connected vehicles.
[00:14:56.549]Alright, something happened, interesting, around 2010,
[00:15:03.077]or at least that's when it became public,
[00:15:05.205]that Google was working on a self-driving car.
[00:15:09.566]The Google self-driving car is direct lineage
[00:15:12.267]from the Stanford vehicle
[00:15:15.290]that won the DARPA Grand Challenge,
[00:15:19.857]well actually, personnel who worked on
[00:15:22.654]the Google self-driving car I think came from both Stanford
[00:15:25.484]and Carnegie-Mellon.
[00:15:30.751]I include this one because I believe
[00:15:33.309]it was really a disruptive moment for the industry,
[00:15:38.597]and it really got a lot of the other companies
[00:15:41.222]and the government thinking about this,
[00:15:45.553]starting to develop their own
[00:15:47.810]to the point that now we see Mercedes, and BMW,
[00:15:51.266]and many other car companies planning
[00:15:56.661]for their own autonomous vehicles.
[00:16:03.524]Around 2012 is when NHTSA really got serious
[00:16:07.100]about looking at autonomous vehicles, automated vehicles,
[00:16:13.930]and started to put together their automation program,
[00:16:18.359]and they're interested in things like the licensing,
[00:16:20.661]testing, regulating, the cybersecurity
[00:16:24.305]is a big concern for them,
[00:16:28.290]and what they did, one of the things they did
[00:16:30.802]was they wrote a letter recommending,
[00:16:37.315]laying all this out and recommending that for now,
[00:16:40.412]states can only legalize automated vehicles
[00:16:44.087]for testing purposes, research and testing purposes.
[00:16:48.924]So NHTSA feels, at least for the time being,
[00:16:52.387]that they're not ready for prime time,
[00:16:54.636]but they obviously see the promise
[00:16:58.100]and want to see the research continue.
[00:17:03.944]NHTSA also has proposed their own taxonomy
[00:17:07.644]for levels of automation,
[00:17:10.274]and I have that shown here in a table,
[00:17:12.892]0-4, where 0 is no automation, 4 is full driving.
[00:17:19.925]So the way this works is level one would be
[00:17:23.004]very specific functions like cruise control,
[00:17:26.716]active warning systems that will engage the break,
[00:17:31.802]or possibly the steering.
[00:17:34.436]When you get to level two is where you start combining these
[00:17:37.454]and the driver can take their hands and their feet
[00:17:41.114]off the controls, but the driver maybe
[00:17:46.132]is still going to be required to monitor the road, detects,
[00:17:51.113]keep track of all the different vehicles,
[00:17:53.308]and signs, and obstacles and objects,
[00:17:58.028]and they may need to take back manual control very quickly.
[00:18:03.053]It's not 'til you get to level three
[00:18:05.030]that you're really allowing the driver to disengage
[00:18:07.460]from the driving task, and perhaps read a book,
[00:18:12.263]watch a movie, even take a nap,
[00:18:15.685]and that's where I think the Google car is now,
[00:18:18.960]although their stated goal is full automation.
[00:18:23.216]And full automation, I'll talk about PRTs in a few slides,
[00:18:27.127]other than that we really don't have anything
[00:18:29.497]at level four right now.
[00:18:31.467]And in the last column is really a heuristic
[00:18:34.696]that's not part of the definition,
[00:18:37.624]but it helps you understand how much vigilance is required
[00:18:42.472]on the part of the driver.
[00:18:44.238]In level one you need to be able to take control
[00:18:47.190]almost instantly, in level two you expect some fair warning,
[00:18:51.900]but you need to be vigilant,
[00:18:53.828]and you need to be monitoring the automation.
[00:18:56.741]In level three you might be waking up from a nap
[00:19:00.880]and need some time to adjust before you can take control,
[00:19:06.350]so these are some good guidelines
[00:19:09.437]to help you understand those levels.
[00:19:12.161]And what's gonna happen in the future,
[00:19:15.240]nobody knows for sure, but some of the vision
[00:19:21.857]might include things light very light vehicles,
[00:19:24.702]because in a crashless environment
[00:19:26.551]we can start reducing the mass,
[00:19:28.098]and that begins what's called a virtuous cycle
[00:19:31.607]where reducing the mass means
[00:19:33.980]that we can downsize the engine,
[00:19:36.228]we can reduce the fuel tank,
[00:19:38.002]we can make the drivetrain brakes and tires all smaller,
[00:19:43.208]and as a result we can reduce the mass some more,
[00:19:47.170]so it becomes very beneficial.
[00:19:53.343]It might have impacts for car ownership in the future
[00:19:58.474]where these autonomous vehicles act more like taxis
[00:20:01.677]rather than things that you own,
[00:20:03.899]and you can dial them up, and they'll come pick you up,
[00:20:07.189]and drop you off and then go pick up somebody else.
[00:20:11.762]So these things are obviously further out
[00:20:14.483]and depend on some good market penetration
[00:20:17.719]and possibly changes in the way that we design
[00:20:21.319]our cities and our roads,
[00:20:23.499]but they're out there, and it's kind of exciting.
[00:20:29.114]So that's my whirlwind tour of the field,
[00:20:33.758]I'm wanna talk now about some of the state of the art,
[00:20:37.738]and there's two ways we can look at it;
[00:20:39.625]one is from a bottom-up approach of developing these,
[00:20:43.403]and the other is more of a top-down approach.
[00:20:54.101]The Advanced Driver Assistance Systems
[00:20:56.644]really represents a bottom-up approach
[00:20:58.704]in that they start as very narrow and specific systems,
[00:21:02.862]even back in the 90s as part of that NHTSA program,
[00:21:10.378]so this chart shows some of the systems
[00:21:13.867]that were commercialized since that time.
[00:21:17.029]This was done in 2011, and unfortunately
[00:21:20.614]it's horribly out of date already.
[00:21:23.768]Today's landscape looks a little bit more like this
[00:21:27.412]where we have a lot more ADAS systems,
[00:21:31.035]we have a lot more complicated systems
[00:21:34.559]that are starting to get into level two automation.
[00:21:42.799]This figure is showing that graphically,
[00:21:48.957]and on the axes I have Action and Perception,
[00:21:52.801]so what that means is every autonomous agent,
[00:21:56.483]whether it be a robot, or automated car, or a thermostat,
[00:22:02.783]really needs to do two things;
[00:22:05.809]it needs to perceive its environment,
[00:22:07.522]and it needs to act on its environment.
[00:22:11.001]For a long time cars have been getting better
[00:22:15.378]in the perception, we've been adding sensors,
[00:22:19.324]we've been starting to add cameras, and radar,
[00:22:23.149]and ultrasonic, and now we're getting into LIDAR and GPS,
[00:22:27.650]and I'll talk about all the different sensory types later.
[00:22:31.344]So cars have been becoming able to perceive
[00:22:36.769]their environment much more, and as a result,
[00:22:40.256]have become more capable of acting on their environment.
[00:22:48.307]Early example would be cruise control,
[00:22:50.111]but now we're getting into active warning systems,
[00:22:53.197]and adaptive cruise control, and lane keeping systems,
[00:22:58.743]and as we get higher, higher in the levels,
[00:23:01.384]we're starting to combine those systems
[00:23:03.927]into more complicated forms like Highway Pilot
[00:23:07.059]or other combinations, and ultimately an autonomous car
[00:23:12.349]will do almost all of the functions of the human driver.
[00:23:19.223]From the other end we have a top-down approach,
[00:23:22.792]and it's interesting that level four automation
[00:23:30.044]has been with us for actually quite a long time
[00:23:34.217]in these Personal Rapid Transit vehicles,
[00:23:36.353]the earliest of which was in Morgantown, West Virginia,
[00:23:39.922]since 1975.
[00:23:42.246]And yes, there are compromises made,
[00:23:45.815]like they use guideways, they run at slower speeds,
[00:23:50.041]but they don't even have controls,
[00:23:52.924]so they're fully autonomous and there's no operator
[00:23:58.244]in the cars at all.
[00:24:00.017]And so it's very interesting technology,
[00:24:03.496]and people are still working at the field from that end.
[00:24:08.408]There are several other examples,
[00:24:13.162]Masdar City is a very ambitious effort
[00:24:17.319]to create a highly technologically advanced city,
[00:24:20.742]unfortunately their system, their PRT system, is on hold
[00:24:26.809]due to some budget cuts, I think.
[00:24:29.925]But Europe has a program called City Mobil and City Mobil 2,
[00:24:33.683]of which the London Heathrow system is a part,
[00:24:36.656]and they have other systems that they're working on,
[00:24:40.542]and what's happening is these PRTs are evolving
[00:24:44.579]so that they're leading the guideways,
[00:24:49.137]driving on plain sidewalks or roads,
[00:24:53.082]still at fairly low speeds,
[00:24:56.793]but they're starting to overcome
[00:24:58.938]some of the early criticisms of the technology.
[00:25:02.025]One was that they just wouldn't scale well
[00:25:04.937]because as you add more, and more, and more cars
[00:25:07.526]you just have maintenance issues and repair issues.
[00:25:15.276]But I think that things are moving in that direction,
[00:25:17.950]and ultimately we're gonna see these top-down
[00:25:19.897]and bottom-up approaches meet in the middle somewhere.
[00:25:25.617]So let's get into a little bit of the technology
[00:25:29.533]that goes into these vehicles,
[00:25:31.744]this is a summary of some of the sensors.
[00:25:38.370]Very conspicuous on the Google car,
[00:25:40.369]is that spinner LIDAR on top,
[00:25:44.263]and it gives you a 360-degree view
[00:25:46.602]of the environment around you.
[00:25:48.700]It was also the most expensive sensor,
[00:25:51.767]and still is one of the most expensive sensors.
[00:25:54.831]The Google car might, well, I heard it cost around
[00:25:58.970]maybe 150K, and the LIDAR that they used at that time
[00:26:03.339]was about half of that cost.
[00:26:06.870]But now LIDAR is in its third and fourth generation,
[00:26:09.854]the cost is coming down dramatically,
[00:26:12.661]and you have these inconspicuous LIDAR systems
[00:26:17.121]that don't have 360 degree field of view, they're narrower,
[00:26:22.200]but they fit on your bumper, on the side of the car,
[00:26:27.957]rather like the radar systems,
[00:26:30.447]and so they don't stick out so much.
[00:26:34.823]GPS and Inertial Measurement Units
[00:26:40.128]have been on cars for a while,
[00:26:43.043]and it's a sensor that gives you a really good estimate
[00:26:46.433]of your location.
[00:26:48.553]Radar has also been used for a while,
[00:26:51.813]and people have different applications
[00:26:54.839]for which they use long-range radar and short-range radar.
[00:26:59.634]There's a couple of different frequency bands
[00:27:02.411]that those operate at, but long-range radar
[00:27:06.010]can detect vehicles up to maybe 400 meters,
[00:27:12.402]and do things like adaptive cruise control.
[00:27:16.756]One of the interesting sensors that's being added is cameras
[00:27:22.272]because it gives you such a different view of the world
[00:27:25.524]than these others, and you can start to look at things
[00:27:28.882]like color, and lighting, and lines on the road,
[00:27:36.615]and it gives you more cues with which you can begin
[00:27:40.034]to classify the objects that you see, that the car sees,
[00:27:43.743]as other cars, or pedestrians, or trucks, or signs,
[00:27:48.074]or what have you.
[00:27:50.598]Then DSRC is the sensor for connected vehicles,
[00:27:54.492]one of the sensor types for connected vehicles
[00:27:56.827]that does networking, and digital maps is something
[00:28:01.448]that people may not think of as a sensor,
[00:28:03.696]but it definitely is.
[00:28:05.364]It's simply a sensor where a lot of the computational effort
[00:28:09.182]has been done by somebody else at a previous time
[00:28:13.333]and you're just accessing it.
[00:28:18.487]What do those sensors do for you?
[00:28:20.343]Well, they give you a view of the world,
[00:28:22.780]they answer the question where am I in the world,
[00:28:26.070]and that problem is called localization.
[00:28:29.466]They also tell you where are other objects in the world
[00:28:35.593]with which I must interact,
[00:28:37.397]and that is the object detection problem.
[00:28:43.683]A car that has multiple sensor types
[00:28:47.509]needs to be able to fuse all of that data together,
[00:28:51.742]it is a difficult problem, but the technology
[00:28:56.375]that people have used to do that has really enabled
[00:29:00.027]the development of autonomous vehicles
[00:29:03.144]over the last few years.
[00:29:06.902]There we go.
[00:29:11.145]One of the keys to doing this has been the use
[00:29:13.527]of probabilistic methods, such as basing and reasoning,
[00:29:18.692]particle filters, Kalman filters, Hidden Markov Models,
[00:29:22.211]because the world is messy, there are uneven edges,
[00:29:26.482]lane lines are fading, people are unpredictable,
[00:29:32.051]and so it's simply not possible to do hard thresholds
[00:29:36.541]that allow you to go one way or the other,
[00:29:41.106]so we work with probabilities, and it's been very effective
[00:29:47.223]for this field.
[00:29:49.485]Along those lines, I'm just gonna put a quick plug in
[00:29:52.149]for some online courses if you're interested
[00:29:55.130]in learning more about that.
[00:29:56.963]I've taken both of these,
[00:29:59.665]one was offered at Stanford originally,
[00:30:02.059]but now is at Udacity, and they're both very interesting
[00:30:05.920]for learning about the technology of autonomous cars.
[00:30:13.096]Digital maps I've already talked about,
[00:30:15.669]what happens when you don't have a digital map?
[00:30:18.937]Well, there are techniques to simultaneously create a map
[00:30:23.947]and then use the map to help you locate yourself,
[00:30:27.049]and that's Simultaneous Localization and Mapping.
[00:30:33.176]Parking lots, there are lots of areas
[00:30:34.934]that you would need to use a strategy like that,
[00:30:37.862]and then algorithms have been around for decades
[00:30:43.725]that are really well-known, like A*,
[00:30:47.053]and A* has been developed, and expanded,
[00:30:49.868]and added onto a lot, to deal with other specific problems
[00:30:55.112]of pathfinding, so that's a fairly well-developed area.
[00:31:02.560]One of the issues with maps is
[00:31:04.371]you must find the proper correspondence
[00:31:06.537]between the digital map and what your sensor's telling you,
[00:31:10.604]so a GPS has a problem because you may be off
[00:31:14.287]by up to a meter, and that's important
[00:31:16.777]because they won't be able to tell you necessarily
[00:31:20.663]which lane you're in on the road,
[00:31:23.591]so the correspondence problem is an important one to solve.
[00:31:27.756]Let's talk about some of the challenges of automation,
[00:31:30.624]one of the primary ones is still weather.
[00:31:33.567]We don't yet know how we're gonna put an autonomous car
[00:31:39.309]on a road in the middle of winter
[00:31:41.928]in Minnesota, for instance.
[00:31:44.886]Snow and things that obscure the air and obscure the road
[00:31:49.655]are just still a big problem.
[00:31:52.371]Connected vehicle technology may be a savior
[00:31:55.759]in that you'll be able to communicate
[00:31:57.253]with the infrastructure,
[00:32:01.964]but this can obscure also sensors on your car,
[00:32:06.790]you can't assume that everyone's
[00:32:08.062]gonna clean their car off, right?
[00:32:10.258]So this is still a very challenging area.
[00:32:15.061]Another one is testing and certification of these devices.
[00:32:20.572]Testing and certification is a difficult problem,
[00:32:22.957]we have the NCAP program
[00:32:25.952]for some of these new collision warning systems
[00:32:30.239]that is able to educate and give consumers
[00:32:35.991]sort of a good picture of what they do
[00:32:41.742]and how well they work, and you get your five-star rating,
[00:32:46.013]but manufacturers have to test NCAPs,
[00:32:53.358]NCAP tests, N-C-A-P, I think that's what your question was.
[00:33:05.867]Let me give you an example of what I mean here.
[00:33:09.539]With some of these systems, systems that we have today,
[00:33:12.505]you need to be able to test,
[00:33:14.188]and normally the manufacturer does this,
[00:33:17.259]for logic failures, sensor failures,
[00:33:21.500]we need to figure out how to reduce noise in their signals
[00:33:24.454]with Kalman filters, they need to be able to measure
[00:33:29.223]false positives, like there are system detecting
[00:33:33.154]bridge abutments, and then stomping on the brakes,
[00:33:38.820]so these are all things that have to be overcome,
[00:33:41.862]and people have done a good job on overcoming them,
[00:33:46.178]but it does add to the testing time.
[00:33:48.826]Now as we start going up the automation scale,
[00:33:53.943]we're starting to add more and more.
[00:33:57.421]We're adding these probabilistic methods,
[00:33:59.821]like particle filters,
[00:34:02.583]we're dealing with data fusion problems,
[00:34:04.469]we're dealing with enormous amounts of data
[00:34:06.929]from cameras and video, and so the number of test cases
[00:34:12.144]are starting to increase exponentially.
[00:34:15.902]I heard one auto manufacturer say
[00:34:18.256]that just for adaptive cruise control
[00:34:20.361]they were able to identify about 3,000 distinct tests
[00:34:24.557]that they needed to cover,
[00:34:29.356]so it really gets crazy.
[00:34:31.929]Going up even further you start to add in decision-making
[00:34:36.872]and path-planning, digital maps,
[00:34:41.739]all the range of speeds so that they'll work in parking lots
[00:34:45.603]and on the highways, and it just explodes
[00:34:49.666]the number of test cases even more.
[00:34:51.945]So this challenge is gonna be significant
[00:34:55.906]before we see a lot of these vehicles on the road.
[00:35:02.984]One of the human factors issues
[00:35:05.625]that's being considered by NHTSA,
[00:35:09.010]first, and they even started tackling this,
[00:35:13.480]well it was the first thing they started looking at
[00:35:16.531]in their automation program, was transfer of control.
[00:35:21.920]And this answers the very fundamental question
[00:35:24.316]of how will the driver give control to the vehicle,
[00:35:28.806]how will the driver take control back from the vehicle?
[00:35:32.850]And let me just show you a little video here,
[00:35:39.363]and this is a very simple video,
[00:35:43.300]what it shows you is a car pulling onto a highway,
[00:35:47.583]pulling in behind a platoon, and then the automation
[00:35:52.043]will first take control of the pedal, the speed,
[00:35:55.921]and then as he settles in behind the platoon
[00:35:58.588]the automation will take control of the steering.
[00:36:02.312]But what's not shown in this video
[00:36:05.263]is what would be the user interface that's required
[00:36:10.790]to inform the driver when they can give up control?
[00:36:15.159]How will this user interface prevent mode confusion?
[00:36:21.339]How will this user interface help the driver
[00:36:25.293]take back control, and is it best to actually do it
[00:36:29.281]in a staged way where you do first pedals then steering,
[00:36:32.714]or is it best to do it all at once?
[00:36:34.895]These are all very fundamental questions
[00:36:39.310]that need to be addressed.
[00:36:42.064]Legality is another challenge of automated vehicles,
[00:36:46.244]there are some states who have paved the way
[00:36:49.142]with early laws, and generally we think
[00:36:54.054]that automated vehicles are legal in most cases,
[00:36:59.140]but a lot of things still need to be decided
[00:37:01.306]such as liability.
[00:37:03.298]If you're not controlling the vehicle
[00:37:04.785]and you get in an accident,
[00:37:06.558]then should you be held responsible
[00:37:08.475]or should the manufacturer be held responsible,
[00:37:11.561]and how much will that add to the price of the car?
[00:37:16.972]Security is another issue that's really important,
[00:37:22.722]it turns out though, that security has always been
[00:37:27.823]somewhat of an issue, people have been able
[00:37:29.649]to hack into cars for a while,
[00:37:32.267]and they've demonstrated this,
[00:37:36.538]and there's many different, all of these things listed
[00:37:38.644]are different entry points that you can use
[00:37:42.039]to hack into a car.
[00:37:45.096]The issue is, before automation
[00:37:48.091]there just wasn't a lot of motivation to do it,
[00:37:50.944]'cause what could you really do?
[00:37:53.642]You could maybe spy on the person,
[00:37:56.441]you could maybe mess with their instruments a little bit,
[00:37:59.990]but now that we're adding automation into these vehicles
[00:38:03.371]a hacker would be able to possibly take control
[00:38:07.294]of the brakes, or run you off the road,
[00:38:10.728]so it's becoming more of a concern.
[00:38:12.562]And with connected vehicles people can do this remotely now,
[00:38:19.564]if these concerns are not taken care of.
[00:38:23.398]As the vehicles get more complex,
[00:38:25.850]they start to add more and more networks internally,
[00:38:29.623]CAN bus has been a popular one,
[00:38:32.068]but there are things that are gonna replace CAN bus
[00:38:34.294]that are even more data intensive,
[00:38:37.494]and more computationally demanding,
[00:38:45.568]so we need to worry about the security of all of these.
[00:38:51.537]Privacy is also a concern,
[00:38:57.847]the black boxes that some cars have,
[00:39:00.437]and that will be required at some point on all vehicles,
[00:39:03.946]are things that people can be concerned about,
[00:39:06.685]but can also help investigators to determine
[00:39:12.050]the cause of an accident, and whether the automation
[00:39:14.963]was really at fault.
[00:39:17.144]The problem is that they record data
[00:39:18.777]that may be considered personal, or,
[00:39:24.086]they just record a lot of data.
[00:39:26.598]Network traffic, if it's identifiable, is a concern,
[00:39:33.201]even if it's de-identified there are ways
[00:39:35.948]to associate data with one car or another by analyzing it.
[00:39:43.886]And as we start to add more and more data
[00:39:46.784]that's being collected, this concern gets greater,
[00:39:51.002]so the Google car, because it records video and LIDAR,
[00:39:55.011]is generating almost a gigabyte per second.
[00:39:58.128]Not that all of that is stored to disk,
[00:40:00.273]but it is generated.
[00:40:04.833]One of the strategies for dealing with privacy
[00:40:07.255]is a concept called Privacy By Design,
[00:40:10.681]and I think that the connected vehicles
[00:40:13.284]is a good example of a program that has taken these things
[00:40:17.261]into account from the very beginning,
[00:40:19.706]and they've designed a network to be secure
[00:40:26.169]and have respect for user privacy,
[00:40:30.319]so I'm somewhat optimistic
[00:40:33.394]about the connected vehicles issue.
[00:40:37.461]Now I want to make a couple of points
[00:40:39.147]that I pulled from my research.
[00:40:42.643]I'm gonna start with a little case study
[00:40:47.205]of an application called Autonomous Intersections,
[00:40:50.924]and I don't do it to pick on this application necessarily,
[00:40:54.932]but I wanna look at it because I think it's the tip
[00:40:57.173]of a very large iceberg of considerations
[00:41:03.036]that we may not be thinking about,
[00:41:05.466]and of unintended consequences.
[00:41:08.537]So there's a group at University of Texas at Austin
[00:41:12.295]who's been developing this for several years,
[00:41:15.023]and I think it's very exciting,
[00:41:17.077]it is a little bit further out in the future,
[00:41:20.082]but the idea is that you have an intersection
[00:41:23.137]that vehicles can communicate with and reserve their place
[00:41:28.695]to go through it, and when they do that,
[00:41:32.468]the timing can be managed
[00:41:34.272]such that you don't have to slow down
[00:41:35.940]or you don't have to stop, so no more traffic light.
[00:41:43.119]This can be engineered so it's completely safe,
[00:41:46.280]but if I put my human factors hat on I have to ask
[00:41:50.989]how that will be perceived by the driver,
[00:41:54.702]so I'm gonna show exactly the same scenario
[00:41:57.207]from the point of view of the driver going through.
[00:42:04.444]These are some of the scenarios, by the way,
[00:42:06.156]that my students were able to develop,
[00:42:09.024]and I call them automation scenarios.
[00:42:13.502]Alright, did anybody flinch?
[00:42:18.928]That is somewhat disconcerting.
[00:42:25.055]One of the reasons that it's a problem
[00:42:27.221]is because people are not always good at estimating
[00:42:30.730]time to collision.
[00:42:33.114]When the time to collision is small they're pretty good,
[00:42:35.552]but when it's three, or four, or five seconds out,
[00:42:40.800]people are not, they tend to underestimate it,
[00:42:43.825]and so because of that,
[00:42:45.116]as you're approaching the intersection
[00:42:46.625]it's just impossible for you to determine
[00:42:51.760]if there's a gap there that you can shoot, right?
[00:42:55.080]So you really have to rely on the automation,
[00:42:57.683]and maybe it's better if you just close your eyes,
[00:43:00.566]so we really need to think about things like that.
[00:43:04.528]And as I said, that's the tip of a large iceberg,
[00:43:08.163]there are issues along those lines
[00:43:11.043]that we need to think of, even today,
[00:43:13.586]as we're dealing with level two and level three vehicles.
[00:43:17.684]So speaking of levels, what's the trouble with levels?
[00:43:23.675]Well we have this nice taxonomy,
[00:43:26.633]I'm going to be wrapping up here
[00:43:27.882]in just a second for questions,
[00:43:30.171]but we have this nice taxonomy that
[00:43:33.386]NHTSA has developed one, SAE has developed one,
[00:43:37.924]there are other examples of taxonomies
[00:43:40.807]for levels of automation
[00:43:42.384]that have been provided in the past.
[00:43:45.953]The problem with all of them is that,
[00:43:47.840]one problem is that they're not a road map
[00:43:50.469]for how this technology will be introduced,
[00:43:56.438]or should be introduced, into the market,
[00:44:00.631]they're just a definition of what the different levels are.
[00:44:05.052]Neither are these levels design guidelines,
[00:44:08.606]they don't necessarily help you at all
[00:44:11.413]design an effective driver vehicle interface, or HMI,
[00:44:16.982]in fact, they might potentially discourage you
[00:44:20.974]from coming up with helpful ideas
[00:44:22.792]like adaptive automation strategy.
[00:44:26.686]Adaptive automation has been put forward
[00:44:29.802]as one solution to the problem
[00:44:35.243]of drivers losing vigilance with automation,
[00:44:39.265]and so an adaptive automation strategy
[00:44:41.959]might jump from one, to two, to three, four,
[00:44:46.290]and back again in order to keep the driver engaged
[00:44:49.958]and optimally participating in the driving task,
[00:44:56.613]but you don't really get that from just looking at this.
[00:45:02.092]Let me give you another example of what I mean.
[00:45:05.393]The levels are clearly presented in terms of,
[00:45:10.471]they follow the technological profile.
[00:45:15.534]The lowest technological challenge is no automation,
[00:45:20.537]and the technological challenges scale
[00:45:24.664]all the way up into fully autonomous.
[00:45:28.565]The problem with that is that people may interpret
[00:45:32.972]that as a road map, that we're gonna go through level one,
[00:45:35.967]level two, level three, level four.
[00:45:38.669]If you look at it from a different perspective, like legal,
[00:45:43.091]you see a different picture,
[00:45:45.528]because from a legal perspective it's really level three
[00:45:50.558]that's the nightmare
[00:45:53.651]because you are allowing the person to
[00:45:58.956]disengage from the driving task,
[00:46:01.461]but still requiring them to take back control at some point,
[00:46:08.052]and so you have all these liability issues that come in.
[00:46:15.553]Let's put on a different hat and look at it
[00:46:17.470]from a human factors perspective.
[00:46:19.869]This is a problem that we're dealing with now,
[00:46:22.352]and NHTSA's well aware, which is why they are tackling
[00:46:26.887]human factors first.
[00:46:29.773]By level two you are allowing the person
[00:46:35.093]to take their hands and their feet off the controls,
[00:46:39.327]but you're still requiring them to be absolutely vigilant
[00:46:42.582]in monitoring the automation,
[00:46:44.137]and if there's anything that we know
[00:46:46.227]from automation research over the past couple of decades
[00:46:49.706]is that monitoring automation passively
[00:46:53.818]is not something that people are generally very good at,
[00:46:56.799]they become complacent,
[00:46:59.885]and so level two is probably the worst situation
[00:47:04.488]from the human factors perspective.
[00:47:07.167]So you need to think about this holistically,
[00:47:10.125]and my fear is that we are actually going to
[00:47:14.947]really progress through these levels rather linearly,
[00:47:19.029]and we're gonna have to deal
[00:47:20.327]with those human factors challenges,
[00:47:22.463]and with those legal challenges.
[00:47:27.949]So this is my last slide, the future, we think,
[00:47:32.363]will be autonomous vehicles, but when will it come?
[00:47:35.231]Some people think five years,
[00:47:36.929]some people think 30 years or more.
[00:47:44.980]I would be interested, if you wanna take an informal poll,
[00:47:48.843]if you could maybe type in what you think will be
[00:47:53.106]the number of years until we start seeing
[00:47:54.887]autonomous vehicles on the road in any significant numbers.
[00:48:04.496]So thank you very much for your attention,
[00:48:07.115]I would really look forward
[00:48:08.284]to hearing some of your questions.
[00:48:16.389]Chris, one of the questions was,
[00:48:19.551]what about pedestrians and bicyclists,
[00:48:22.546]would they need to wear a transponder
[00:48:25.066]to survive in this environment?
[00:48:26.968]Okay, that's a great question.
[00:48:32.019]Well, that problem is being considered
[00:48:36.946]even from an ADAS perspective,
[00:48:39.183]so the new collisional warning systems
[00:48:43.070]are beginning to look at pedestrians,
[00:48:45.175]and smaller objects like that, and do autonomous braking,
[00:48:52.283]and that's being done today with radar.
[00:48:57.256]The introduction of cameras is going to make that technology
[00:49:01.414]more robust and more powerful,
[00:49:06.666]I would say, as it starts to classify the objects
[00:49:11.097]into pedestrians, or bicyclists,
[00:49:14.646]and once it does that, if it knows
[00:49:16.634]that it's going to be a pedestrian, for instance,
[00:49:20.995]then it can begin to make some assumptions
[00:49:23.221]about the movement patterns of that object.
[00:49:25.613]It knows it's not gonna move
[00:49:26.889]at 30 miles an hour, for instance.
[00:49:31.284]I think it's a challenge because the problem is
[00:49:35.555]you really can't afford to make any mistakes, can you?
[00:49:40.067]The first time that there is an incident
[00:49:42.761]between an autonomous vehicle
[00:49:44.369]and a pedestrian or a bicyclist,
[00:49:47.047]we don't really know what's gonna come out of that.
[00:49:49.518]But I think that the technology is there,
[00:49:53.412]and it's gonna get better to deal with those.
[00:50:23.083]If anyone has any questions
[00:50:24.430]please submit it on the chat forum.
[00:50:38.729]How much participation,
[00:50:40.993]yeah, so-- One of the questions...
[00:50:46.263]The question was how much participation
[00:50:47.892]are you expecting a driver
[00:50:52.341]to get involved in, and how do we determine
[00:50:54.289]the responsibility when an accident occurs?
[00:50:56.873]Is this a question about liability,
[00:51:00.064]in case there's an accident?
[00:51:08.940]I would think so. That's a really
[00:51:10.058]interesting question, and I will admit up front
[00:51:12.971]that I'm not an expert on the legal side of things,
[00:51:17.370]but from what I've read, from a legal framework,
[00:51:22.439]one of the issues is how autonomous vehicles
[00:51:25.240]will be treated.
[00:51:28.982]And one of the ideas about this
[00:51:36.777]is that you could think of autonomous vehicles
[00:51:38.883]a little bit more like elevators
[00:51:41.750]than you would something like trains,
[00:51:47.256]and what that means for the legal framework of liability,
[00:51:55.219]I'm not entirely sure of,
[00:51:57.844]but I think it's safe to say that there will be
[00:52:01.263]some transfer of liability from the driver
[00:52:03.723]to the manufacturer, that's definitely going to happen.
[00:52:13.781]I would hope that we will be able to determine,
[00:52:18.516]through the forensic analysis, exactly what was the cause
[00:52:24.055]of any particular accident.
[00:52:29.586]Yeah, I don't know exactly, it'll be interesting
[00:52:32.054]the first time that there is some kind of accident,
[00:52:34.642]exactly what's gonna happen, because some people have said
[00:52:37.894]that it's gonna be the end of the field,
[00:52:40.984]the first time there's a big accident
[00:52:43.249]involving an autonomous vehicle,
[00:52:45.255]then society will reject them and that'll be the end.
[00:52:48.605]And I'm not that pessimistic about it,
[00:52:51.178]but I don't exactly know what's gonna happen.
[00:52:57.362]There was one question regarding,
[00:52:59.075]does California or Nevada, that have allowed Google car,
[00:53:02.501]have had any problems in terms of crashes?
[00:53:06.549]No, the Google car has not had any crashes.
[00:53:10.977]There have been, I think, a small handful of incidents,
[00:53:17.293]but I think that even those incidents happened
[00:53:21.387]while it was in manual control and not autonomous control.
[00:53:25.461]So I think they've been very successful
[00:53:28.155]at managing their research
[00:53:32.162]such that it's been fairly safe on the road.
[00:53:35.716]Actually I had a chance to drive, or ride in the Google car
[00:53:38.523]last summer on the highway out there,
[00:53:41.329]and it did a great job.
[00:53:47.699]The other question is,
[00:53:48.574]is there any pushback from the automakers
[00:53:51.276]since ownership would be less in this plan?
[00:53:56.790]I haven't heard of any.
[00:53:59.582]The ownership issue is further out in the future,
[00:54:04.865]and I think that the concern of automakers right now
[00:54:09.181]is more one of being left behind.
[00:54:14.644]They want to get their offerings out there
[00:54:17.912]and be sure that they can begin to fill these niches
[00:54:23.156]and the demand that there might be for automation.
[00:54:30.042]Yeah, I...
[00:54:34.447]I think that they're probably looking at the ownership issue
[00:54:37.257]as something that they can try to shape themselves
[00:54:41.638]through their influence, and through policy making,
[00:54:46.067]so I'm sure that they will have a handle,
[00:54:48.489]they'll have an interest in the issue
[00:54:50.051]and they'll try to affect it as much as they can,
[00:54:53.752]but it hasn't been something that's been causing them
[00:54:56.936]to push against automation right now.
[00:55:06.648]One question is what do you think
[00:55:08.463]the role federal government is in regulating and funding
[00:55:11.945]the development of this technology?
[00:55:14.393]It'll be significant of course.
[00:55:17.683]NHTSA is taking an approach right now
[00:55:21.886]that is one of research, and they're not rushing into
[00:55:27.048]the regulation side of things, which I think is good
[00:55:31.636]because there's a lot of innovation
[00:55:33.960]that has yet to be done by Google,
[00:55:38.653]and by the other manufacturers.
[00:55:43.528]But clearly there is gonna be a tremendous need
[00:55:46.655]for regulation in the future.
[00:55:53.210]The issues that you see today with cars
[00:55:57.551]like transfer of knowledge from one car to another,
[00:56:04.946]when these systems are so complicated
[00:56:06.720]and you get confused, that happens to all of us
[00:56:10.538]to some degree, and that's gonna become worse
[00:56:14.889]as you have systems that are more complex,
[00:56:18.201]and they may distract you, you may not understand
[00:56:22.585]what they're doing, and you may just get into confusion
[00:56:28.085]about what mode you're in.
[00:56:30.322]Mode confusion is one of the reasons
[00:56:31.726]we still have airline crashes today,
[00:56:35.883]so if that becomes a cause of car crashes in the future,
[00:56:41.905]it's a serious issue.
[00:56:49.024]About the aviation industry,
[00:56:51.090]is there any lessons to be learned
[00:56:52.547]in terms of what they're doing in legal issues
[00:56:57.889]when a plane crashes on autocruise or something?
[00:57:00.938]Yeah, there certainly are, and the mode confusion issue
[00:57:03.813]I just referred to was one.
[00:57:09.842]I don't know exactly what all we'll pull from aviation,
[00:57:15.781]because there are definitely differences
[00:57:19.905]between the two, and the level of quality assurance
[00:57:25.956]and reliability that aviation needs to meet
[00:57:29.532]is much higher than what the auto industry needs to meet,
[00:57:35.094]and rightly so.
[00:57:37.286]But I think there are definitely lessons to be learned
[00:57:40.453]on the human factors front and on the reliability
[00:57:43.941]and security front.
[00:57:53.047]There is one question regarding
[00:57:55.214]the rate of penetration of autonomous vehicles,
[00:57:58.305]what penetration rate would be a liable option
[00:58:01.455]for the whole system to work?
[00:58:03.258]That's a great question.
[00:58:09.446]The way things are going, we're introducing more and more
[00:58:12.879]these collision warning systems, ADAS systems on the road,
[00:58:18.569]so it's a little bit organic,
[00:58:22.010]and that will continue, I think.
[00:58:25.119]In terms of actually jumping to a much higher level
[00:58:28.854]of automation, I think the nice thing is
[00:58:32.801]you can put those cars on the road
[00:58:34.529]and they can drive in mixed traffic just fine.
[00:58:38.038]They have to be able to follow the rules of the road
[00:58:40.596]and interact with human drivers.
[00:58:45.987]Now, there are opportunities to do things like platooning,
[00:58:51.646]and for that, states and communities
[00:58:56.878]may be looking at managed lane applications,
[00:59:01.516]so we may start to see express lanes, or automated lanes,
[00:59:08.949]I think that's a possibility.
[00:59:13.084]But I think for the most part,
[00:59:17.197]I see this happening gradually and organically,
[00:59:21.377]and I can't really say when we'll see
[00:59:25.271]significant market penetration,
[00:59:27.138]that's the $64,000 question right there.
[00:59:35.484]One question is regarding,
[00:59:36.744]is there any industry-led effort
[00:59:38.600]to streamline the validation process for
[00:59:40.429]automatic driving systems? I'm sorry,
[00:59:41.405]could you repeat that question?
[00:59:46.687]Are there any industry-led efforts
[00:59:48.838]to streamline the validation process
[00:59:51.150]for automated driving systems?
[00:59:57.980]I think that people are thinking about it a lot.
[01:00:01.971]When I was at the TRD summer workshop last summer
[01:00:05.133]there were certainly industry representatives there
[01:00:09.472]who were talking about the issue,
[01:00:12.886]like the example I brought up of all the test cases for ACC.
[01:00:20.991]And the technologies for dealing with things like that
[01:00:25.579]are ones that industry has been developing for some time,
[01:00:31.268]like model-based design, and simulation-based tests.
[01:00:38.701]So I think that there is quite a bit of work
[01:00:43.611]on building out the test structure for the V model,
[01:00:47.890]for verification and validation,
[01:00:51.135]but given the explosion of the number of test cases
[01:00:56.697]that we need to deal with,
[01:00:59.621]I think it is gonna be a challenge
[01:01:01.658]to try to automate these tests
[01:01:04.741]and try to find ways to deal with
[01:01:09.527]just a larger quantity of data.
[01:01:16.032]I think we have time for one last question.
[01:01:18.979]The question is, would the cars all go at the speed limit,
[01:01:22.693]or can they go over?
[01:01:24.730]Or can they go over, yeah.
[01:01:31.703]Yeah, I don't know.
[01:01:37.769](woman speaking softly)
[01:01:39.218]Yeah, right now.
[01:01:40.882]Maybe we should raise the speed limits a little bit.
[01:01:43.983]But certainly if they're in mixed traffic,
[01:01:47.183]drivers are gonna be going above the speed limit,
[01:01:50.971]so should the autonomous vehicle go the speed limit
[01:01:54.117]or should it follow the traffic?
[01:01:56.517]On the one hand, what are the implications
[01:02:00.330]of an autonomous vehicle that doesn't follow
[01:02:02.349]the letter of the law?
[01:02:04.047]That's a serious issue.
[01:02:06.432]But on the other hand, if it's not following
[01:02:09.209]the flow of traffic adequately,
[01:02:11.359]then it'll become frustrating for the driver.
[01:02:14.461]So one of the things that Google saw
[01:02:16.393]in testing their car, as they would come to an intersection,
[01:02:20.882]is that it would wait its turn,
[01:02:23.939]and when it would wait its turn to go
[01:02:25.908]it would never get to go,
[01:02:27.139]because people would just go in front of it.
[01:02:29.574]So they had to actually build behaviors into the Google car
[01:02:33.690]to inch out into traffic
[01:02:36.844]and signal its intention to go,
[01:02:40.942]and then people would give it the right-of-way.
[01:02:43.515]So I think that's a really interesting question,
[01:02:48.784]I don't fully know the answer to it yet.
[01:02:55.955]Thank you very much Chris, for your time,
[01:02:58.301]and the presentation recording will also be posted online,
[01:03:02.232]so thank you all for attending the seminar series.
[01:03:08.216]Thank you.

The screen size you are trying to search captions on is too small!

You can always jump over to MediaHub and check it out there.

Tags:
ntc
matc
nebraska transportation center
mid-america transportation center
autonomous car
chris schwarz
civil engineering
university of iowa
autonomous vehicles
national advanced driving simulator
advanced driver assistance systems
matc lecture series

Comments

0 Comments

2014 MATC Lecture Series - Dr. Chris Schwarz, "Towards Autonomous Vehicles"

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments