In 1991, Congress authorized $650M to make driverless cars a reality (2013) | Hacker News Reader

In 1991, Congress authorized $650M to make driverless cars a reality (2013)

smithsonianmag.com

150 pointsjameslk71 comments

When I was studying computer science at Carnegie Mellon in the '90s, I played bass in a band, and the guitar player was a grad student at the Robotics Institute working on the autonomous vehicle project, the reason CMU was part of the consortium in the article. NavLab 2 was an Army Humvee with a laser scanner mounted on the front, along with some cameras and a bunch of other sensors, and some powerful-at-the-time computers (Sun SPARCstations, IIRC) in back. It was a pretty impressive sight: https://www.cs.cmu.edu/Groups/ahs/images/navlab_1_5_images/n...

It was usually parked in the Field Robotics garage, but if you went out early on a weekend morning, you might catch it driving itself slowly around Schenley Park (which is right next to CMU campus) – without anyone at the wheel, which was really unreal and sci-fi-ish at the time.

I remember one story of a jogger who was surprised to suddenly run across it. The story goes that the he or she screamed and actually fainted. The truck was programmed to stop if it found itself in front of any obstacle, so when the jogger came to, there was the truck looming above.

I have no idea if the story is actually true, but I like it anyway.

For anyone else having trouble loading the image: https://web.archive.org/web/20160920032328/https://www.cs.cm...

That's pretty incredible - they let it run unsupervised? Or did the jogger just faint for a very short time period?

I remember it running autonomously on the little roads in Schenley Park, which were wide just enough for the vehicle but not open to traffic. Unfortunately, I never learned what happened to the jogger.

I heard the same story about the jogger being hit by the autonomous Humvee at CMU. I ended up going somewhere other than CMU, but the onsite interview process included a tour, and that tour mentioned this jogger being 'hit'.

Probably the reason why the jogger screamed and "fainted".

One the problems with government run transportation R&D is that the responsible agencies are fundamentally not in the R&D business. DoT and subordinate agencies like the FAA are regulatory and risk management organizations (with some mandates in inspection, construction and telecommunications). They lack the expertise, structure and know-how to build effective Research and Development programs and to overcome their own institutional risk management culture. For example, the FAA has been trying to roll-out a new air traffic control system for decades and at a cost that regularly far overruns any estimates the Agency can put together.

The DoD, DoE and the Intelligence Agencies have dedicated R&D arms (DARPA, IARPA, ARPA-E, etc.) with long legacies of knowing how to to research and shepherd it into operational development states and/or commercialization. Even other parts of the civilian government has things like the NSF,

The closest the civilian world has for transportation is maybe NASA? But for self-driving cars, the DoT's R&D components are not well organized and seen very much as side-show "Administrations" and are generally new affairs. DOT's version of DARPA is called "RITA" and was formed only in 2005 by strapping together a handful of previous organizations inside of DoT and are focused mostly on, you guessed it, improving safety and studying statistics.

It's no wonder that most of the government research that's actually made progress in this area have all come from DARPA.

I think you're missing a few factors. Most critically, DARPA has some success because they have a customer (the DoD) with specific requirements and a nearly unlimited source of funding compared to any other government initiative. The Defense Department probably wastes more money each year on meandering, unfinishable snake-oil projects than has been spent on driverless car R&D by all entities in all of history.

But it's true that driverless car tech is unlikely to come out of direct government funding. Instead, that money should be going to fund quality education at all levels, and to build proven technologies that would solve many of the same problems the driverless-car pipe dream would only be an inferior version of--ie high speed rail links in the middle distance where it's a better form of transport than cars (driverless or not) and planes.

Driverless cars are directly coming out of DARPA funding ex: https://en.m.wikipedia.org/wiki/DARPA_Grand_Challenge. Google for example can trace it’s program directly back to DARPA.

Companies are more willing to fund after the basic proof of physical possibility shows up. Vs the kind of this might be possible in 30 years projects DARPA gets off the ground.

PS: It’s often just as much about people getting experience working on these problems as it is any specific technology.

Driverless cars aside for a second, government grants are essential to R&D and commericialization. Having learned recently about Small Business Innovation Research (SBIR) and Small Business Technology Transfer (STTR) grants, I can't believe I didn't know about it

Department of Transportation topics for 2018: https://www.sbir.gov/node/1413255

Even if the research isn't done by government itself, it's likely the result of government grants. VC usually step in and reap profits way later

Most of those grants require ridiculous amounts of paperwork, with projections and business plans. That basically forces you to lie and jump through hoops to get through the process.

There are plenty of companies that are experts at getting those funds, which is essentially part of their business model. But it can be a big distraction for a small firm where it would likely be better to find private funding sources.

I worked for such a firm. Basically almost nobody gets to a Phase III which is the commercialization stage. If anyone does get to that stage, they are not usually set up to make a commercial product and generally require some kind of licensing or partnership to make commercialization possibile. This is typically because these businesses are set up to satisfy the bureaucracy of the SBIR award process and not to actually produce a product (in most cases). In another case I worked at a VC funded company that had some government contracts to try to help self fund R&D activities. These activities proved to be orthogonal to our main product focus and became paperwork exercises. I had to fill in a time sheet for years where I put zero hours on it.

I’m thinking about getting into this space — any chance you could share a bit about the experience?

Tempting to claim that Silicon Valley’s methodology is the only way to come under budget.

Even if it runs afoul of basic ethics and safety concerns.

I'd argue that the apparent lack of ethics and safety concerns are what make Silicon Valley more effective in this arena. There's no upside for government agencies (or, more specifically, the bureaucrats that greenlight research proposals) to take on high-risk projects like these, and plenty of downside. SV-types have a ton of potential upside that they weigh against ethics and safety. Their calculus is different, and makes them far more aggressive in advancing their agenda. Imagine a government-sponsored version of Uber, for a moment. How fast would they have grown? Would they even have gotten off the ground or just folded at the first sign of resistance from current stakeholders (taxis and limos). Not to say that SV-types shouldn't be held accountable, and they're certainly capable of wasting money, but man do they get stuff done in a hurry (for better or worse ...)

Yeah it's a huge problem the military has such a monopoly on this stuff, and civilian agencies are so cut out. I hope I live to see this reversed.

We invented horseless auto mobile but required continuous human attention for steering and managing the machine. With a horse indeed we already had semi-driverless operation in some scenarios.

We are reinventing the "horse" - sometimes I think.

And I believe less consumption of fossil fuel

Automobiles consume much less fuel during long-term idle periods.

Unfortunately they also encourage much more use, and thus, less idle periods, long-term or otherwise.

However, automobiles require significantly more resources and energy to produce.

We're reinventing Nature anyway. When we'll have to go very tiny, green we'll have insects/animals. Except with a programmatic interface. Not sure which is best.

Maybe all of biology is the previous civilization reinventing nature. (Deep thought)

If Henry Ford had asked the people what they wanted, they would've asked for a better horse.

And what a better alternative world that would be...

I often remark that people really did want a faster horse.

Death's robot horse from Hickman's 'East of West', ideally. https://latimesherocomplex.files.wordpress.com/2013/10/easto...

How hard would it be to put an actual horse in control of a car?

You get the speed of a car, and the self driving of the horse (or some other animal).

Here's a brief history tracking the progress of developments in that area. I'm not sure where this tech lies on the Gartner Hype Curve though.

Pigeon controls missile - https://en.wikipedia.org/wiki/Project_Pigeon

Horse drives car - https://www.horseandhound.co.uk/videos/news-videos/buttersco...

Goldfish controls robot - https://www.theregister.co.uk/2017/01/26/robot_goldfish_tank...

edit to add one more as I realised that one of my favorite 'most worrying tech developments' fits right in...

Rat neurons cultured on silicon substrate control robot - https://www.technologyreview.com/s/401756/rat-brained-robot/

At least cars don't produce four tons of manure per year

https://99percentinvisible.org/article/cities-paved-dung-urb...

Worse, they produce 4.75 tons of CO2.

https://www.reference.com/science/much-pollution-average-car...

Surprisingly, the environmental impact of a horse is even higher.

Public transportation or perhaps electric scooters are the only sustainable ones.

walking & cycling.

Electric bikes can be quite enjoyable to ride, the Bafang BBSHD can make it a breeze to go up even steep hills here in Seattle at a swift clip.

The downsides come in the form of risk in transit (your in the weather, and have minimal to moderate protection from the road and others on it) and in securing your newfound transport in between use.

There are very few dry, secure placea to store bikes & scooters, and fewer yet that have an outlet to charge. Shared bikes and scooters mitigate this problem somewhat, but thus far they seem to have serious maintenance issues after relatively short periods outdoors. Who wants to ride a rattly transport device thats missing screws?

Its not so much of an issue if you own the bike. How many people work at a place that has literally nowhere to place a bike inside and no free power points to use. I have always placed my bike next to my desk.

Public transport and electric scooters only make sense in dense urban areas. In these places there's a critical mass of people that want to go to the same set of locations. This premise doesn't hold true in most of the world. Outside of dense urban areas public transport has no benefit from scale and is worse in terms of energy usage than private cars with plenty of other downsides.

Most of the world is not dense but most of the population is. Especially if you look outside of America.

I have been a proponent of electric scooters nonstop since a recent trip to Prague, where they have Lyme. I had a first date riding them around to no particular destination.

Unfortunately, the stupid helmet laws in my city (Melbourne) mean no one will ever use them even if we construct more bike lanes.

From what I heard, scooters can very quickly slam you in to the ground if you hit a hole/rock or if the front wheel turns 90 degrees which many of the company's scooters allow. Bikes seem much safer than scooters to ride helmetless.

Fun fact: I'd you go around older and we'll maintained European cities you'll see little holes with metal bars next to most front doors. Those were to scrape your shoes clean of all the street dirt (mainly horse products) before entering.

Pretty sure horses needed continuous human attention. "Driver" wasn't the operator of a machine, just like "car[riage]" wasn't originally an automobile.

Horses are pretty good at finding back to their stable on their own.

The difference is that a cars don't react unpredictably unlike horses. Spooked horses are very dangerous things and can run off on their own. We certainly wouldn't accept horse brained cars on the road.

a well trained knight's horse i think would be harder to spook than the most of the human drivers we have around.

Technically true but they would be more akin to a severe case of road rage. Those warhorses were kept isolated from others to be kept 'trained' in the right animal frame of mind and essentially meant to act like violent psychopaths - 'working as intended' was to attack anyone except the rider, even when the rider was unhorsed. There is a reason why they were rare, expensive, and why they also had separate general riding horses with much gentler temperaments who could actually share stables.

Romanticizing the past and thinking of it as an ideal that never really was and using it as a model is a rather dangerous pattern throughout history which has never really ended well.

One of the most common pieces of equipment for horses were 'blinkers' or 'blinders' - whose function was to limit their peripheral vision.

> Needless to say, the program didn’t deliver driverless cars and automated highways to Americans. So what was the problem? The legislation didn’t really give the Department of Transportation any direction on how they should go about the research—only that they needed to demonstrate it by 1997. But perhaps the biggest problem was that the legislation never clearly defined what was meant by “fully automated highway system.”

Look a lot more like the problem was that there just wasn't any follow through. Prototypes were made, tests were done, proofs of concept made, but $650M is not enough to overhaul an entire nation's infrastructure, and one shortish strip of highway isn't enough to convince people to go out and buy whole new cars.

I found this article linked to a video someone posted here: https://news.ycombinator.com/item?id=18755243

The video is also pretty interesting: https://www.youtube.com/watch?v=RlZEeIC_2lI

> In the future, which may be 10, 50 years from now, there won't be a driver

@1997

21 years are gone.

> You'll be able to read the newspaper on your way to work.

Yeah. About that...

The opening titles are even better:

> Highways of the future may feature relaxed drivers, talking on the phone, faxing documents

> talking on the phone

Again, they were not completely wrong...

In early 90s, BMW had a working prototype IIRC. Something quite elaborate even though the processing power was limited. Kodak moment.

I believe one of the earliest non-guided (so actually autonomous) self driving cars was the Vamors by Dickmanns in the 80s. IIRC it used one or two racks stuffed with transputers to run the machine vision and control software. Later there was the Euroka project based off that.

> ... achieved in 1995, when Dickmanns´ re-engineered autonomous S-Class Mercedes-Benz took a 1,000 miles (1,600 km) trip from Munich in Bavaria to Copenhagen in Denmark and back, using saccadic computer vision and transputers to react in real time. The robot achieved speeds exceeding 175 kilometres per hour (109 mph) on the German Autobahn, with a mean time between human interventions of 9 kilometres (5.6 mi). In traffic it executed manoeuvres to pass other cars. Despite being a research system without emphasis on long distance reliability, it drove up to 158 kilometres (98 mi) without any human intervention.

I worked the spring and summer of 1991 at Daimler-Benz as an intern on their autonomous vehicle (VITA, a 7-ish meter bus at the time). This was the 2nd generation bus and I think the immediate predecessor to the S-class mentioned in parent comment. We’d sometimes test in Munich or Turin with the other European auto marquees, but mostly tested on an abandoned stretch of Autobahn just outside of Stuttgart.

I worked on the vision system (as in [under the direction and guidance of my boss] wrote all the code to improve frame rate from 5Hz to 10Hz, double the processing resolution (to around 128x96 IIRC), and implemented Kalman filtering for preceding vehicle detection and ranging). System was built in Occam on Inmos transputers. Each was about $15K and roughly the power of a 486 but with fast, reliable message passing. We could auto drive (lane follow and distance keep only) using only vision, but not especially reliably; it would fault and give up every 3-15 minutes, depending on lighting and roadway conditions. Partially wet, drying roadways remained a big problem as I left at the end of the summer.

Really cool internship. Part of https://en.m.wikipedia.org/wiki/Eureka_Prometheus_Project

VITA image: https://goo.gl/images/Cz8ZuZ

I wonder how fast did these companies ran to their archive department to find notes from that project when Google, Tesla et al started to pitch level5 auto-automation.

ps: Also did you know about the pseudo self driving system made in the US (50s) by embedding a radio wire in the road so the car above could stay in lane ?

Didn't know about the radio guidance; will have to google it.(I do know about the 4-course aviation radio nav systems, but am not old enough to have flown them.)

With regards to quickly searching the archives, I suspect all the major manufacturers have had a reasonably steady, perhaps varying in intensity, advanced research group. So, it's probably less searching the archives and rather just asking the grizzled engineers in those groups...

I was watching First Man recently, the movie about the first moon landing.

I'm wondering how much the popularity of computers changed how SpaceX was able to develop its software and properly analyze the data. I guess it made things easier, but to be honest I have no idea how hard NASA worked its computers and systems to use data in the 60.

I always find it hilarious that they test drive autonomous vehicles in Arizona. Never in Tokyo or Amsterdam.

Why? If they don't work in Arizona, they won't work anywhere else.

If AV doesn't work in Tokyo, they may still work in many places.

I think this sort of illustrates just how hard iterative development is for most people to grasp. It's similar to how junior engineers frequently aim for the final finished product on the first iteration whereas most experienced engineers will iterate on development when in unfamiliar spaces.

Second the Amsterdam bit: bikes all around you makes for some very interesting driving & AI would need extensive extra training over there. Self-driving cars will need to be localized for traffic “culture” all around the world.

Or they could not use them because bikes and trams are just fine.

Nevermind that bikes and trams are significantly cheaper, and with the way the infrastructure is designed in many places (plus existing traffic), other modes of transit besides cars can be competitive or significantly faster than driving.

And never anywhere it snows.

Like here? https://www.theverge.com/2018/2/16/17020096/self-driving-car...

They do test in snow areas. Just many fewer of them and with no successful feasible solution. Notice that the road in the video is cleared of snow.

The problem is when the road is covered with snow. This is a very common occurance. So far it requires solutions like pre-existing ground penetrating radar mapping of all roads or some not invented yet solution. There's a team at a Michigan university doing this.

Testing in Tokyo: https://www.theguardian.com/technology/2018/aug/28/driverles...

I rode in one of these. It seemed pretty far from production ready, and the driver had to take over several times. But, I was impressed with the testing philosophy. 2 engineers rode in the car, and the taxi driver had only the responsibility to be ready to take over driving, at all times. And by partnering with a taxi company (which are well respected in Tokyo), they automatically have trust, hailing infrastructure and a way to collect data going forward.

They don’t appear to be operating regularly yet, and the timing of public testing is probably due to a recently failed IPO rather than readiness.

You want a car programmed to drive on the right on Japanese roads?

No. No. No!

This can’t be true!

I like driverless cars. My libertarian friends tell me that nothing the government does is good.

Theeefore this cant possibly be true.

Same thing with the internet.

Despite the sarcasm, this is a good point. Which revolutionary technologies have not been developed with public money?

That the government spends a shitload of money on basically everything under the sun doesn't mean that public investment is always a good idea. The Internet was very largely developed by the private sector; despite public underpinnings of the infrastructure, the private sector would have brought it along in time as well and the result would have been largely the same.

The private sector was the phone company, and their vision was quite different. Without huge amounts of federal funding creating operating system support, tools, and operational practice the future might have looked at lot like CompuServ or AOL. I don’t know whether they still have it but the old displays at Epcot used to be a fascinating warped-mirror alternative history.

Reply on news.ycombinator.com