Legality of DIY Autonomous Vehicle on the Road
Heath Matlock
control system in their car for driving on a public road, right now in
the U.S.?
--
Heath Matlock
+1 256 274 4225
Paul D. Fernhout
> What is the legality of one using his or her own autonomous vehicle
> control system in their car for driving on a public road, right now in
> the U.S.?
Good question. I don't know. I know I was told about eight years ago or so
that people around a certain university were doing it clandestinely in their
own cars. Again, is that legal? Certainly the driver would still be liable
for accidents. Reckless endangerment? So, I don't know. It's a really good
question.
Personally, I've long thought a good system to work on would be one that
controlled a Playstation driving game (Gran Turismo?)
http://en.wikipedia.org/wiki/Gran_Turismo_%28series%29
or some XBox driving game. So, you hook up a separate computer that gets the
images from the game somehow (USB camera? direct video capture?), and
controls the steering, throttle, and other things via a USB cable. Some of
the games are getting so realistic with dust and other effects that it would
be a challenge. So, that is not as impressive as modding your car, but safe,
legal, and probably still fun, and it gets you doing research in vision
processing, path planning, and so on. One might then generalize it to other
vehicle games (snowboarding, ATVs, boats, whatever). And it is probably much
cheaper.
Of course, it might be illegal to use that kind of software in online
head-to-head driving competitions? :-) Maybe people do that already?
--Paul Fernhout
http://www.pdfernhout.net/
http://www.beyondajoblessrecovery.org/
Heath Matlock
<pdfer...@kurtz-fernhout.com> wrote:
>
> Good question. I don't know. I know I was told about eight years ago or so
> that people around a certain university were doing it clandestinely in their
> own cars. Again, is that legal? Certainly the driver would still be liable
> for accidents. Reckless endangerment?
I guess it's legal until made otherwise. To my knowledge, and having
not checked with authorities yet, there isn't a law for this.
> Personally, I've long thought a good system to work on would be one that
> controlled a Playstation driving game (Gran Turismo?)
> http://en.wikipedia.org/wiki/Gran_Turismo_%28series%29
> or some XBox driving game. So, you hook up a separate computer that gets the
> images from the game somehow (USB camera? direct video capture?), and
> controls the steering, throttle, and other things via a USB cable. Some of
> the games are getting so realistic with dust and other effects that it would
> be a challenge. So, that is not as impressive as modding your car, but safe,
> legal, and probably still fun, and it gets you doing research in vision
> processing, path planning, and so on. One might then generalize it to other
> vehicle games (snowboarding, ATVs, boats, whatever). And it is probably much
> cheaper.
I do think that simulation is the way to go first before implementing
a physical system, and it's some simulation software which I saw for
this which made me wonder.
http://spectrum.ieee.org/green-tech/advanced-cars/autonomous-vehicles-complete-darpa-urban-challenge
(November 2007)
"Six of the 11 contenders made it through the entire trial, covering
roughly 55 miles in less than 6 hours. The winner averaged roughly 23
km/h, and maximum speed on the fastest part of the course was 48
km/h."
At that speed, something like the Airpod could use such a system, and
blame its lack of speed on not the computer, but the limits of its
energy source ;-)
http://spectrum.ieee.org/energy/environment/deflating-the-air-car
(November 2009)
Here are some papers which came from the DARPA challenge:
http://graphics.tu-bs.de/people/lipski/projects/carolo/index.html
I went back to find the software which had me wondering in the first
place, turns out I misread the description:
http://sourceforge.net/projects/agvsimulation/
"This project about providing run-time support for developing
Decentralized Autonomic Computing systems (Eclipse Innovation Grand).
Currently the project contains a automatic guided vehicle (AGV)
simulator and an editor and visualisation as eclipse plugin."
And I was curious what they used in the DARPA challenge, but
initially, to me it looks like the software and hardware are all
patented and non-free.
http://www.torctech.com/content/darpa-urban-challenge mentions five
technologies used:
SafeStop™ - A low-cost, multi-level, wireless emergency stop system,
the SafeStop was used by many Urban Challenge teams and has been
adopted as a safety standard in several commercial robotic
applications.
ByWire XGV™ - A Hybrid Escape drive-by-wire platform, the ByWire XGV
makes Odin’s seamless conversion and superior design available to
customers worldwide.
AutonoNav™ - A ruggedized, compact computing module, the AutonoNav
uses advanced Route Planning and Motion Planning software (similar to
that used in the Urban Challenge) to rapidly convert ground and
surface vehicle platforms from drive-by-wire to autonomous navigation.
SimVironment™ - A distributed, scriptable, physics-based 3D simulation
environment, SimVironment reduces software development times and
allows for more complete and efficient testing while increasing safety
and reducing cost.
Autonomous Vehicle Software System - Odin’s proven hybrid
Deliberative/Reactive software architecture and interchangeable
software modules can be easily adapted to custom mobile robotic
applications, from open-pit mine automation to site surveillance and
security.
The first two technologies mentioned on the list don't seem too
difficult, so what I would like to do is take this thread into the
direction of what an autonomous vehicle simulator would require, and
also discuss what the Autonav clone might include.
What confuses me about Autonav is that it states on the site: "similar
to that used in the Urban Challenge", was it not used in the event? At
any rate, something like Kapal ( http://code.google.com/p/kapal/ )
might be similar:
"Kapal is an open source path-planning library written in Python. Its
intended use is algorithms research and navigation tasks in games and
robotics."
What free software GPS software might be used?
The SimVironment clone would require something like Vdrift (see
http://vdrift.net/ and http://en.wikipedia.org/wiki/VDrift ) I'm
guessing.
And the Autonomous Vehicle Software System uses a computer vision
library. An incomplete list of computer vision software:
http://sourceforge.net/projects/opencvlibrary/
http://code.google.com/p/pycam/wiki/PythonComputerVision
http://code.google.com/p/sdpy/
And here's another paper about using Python for CV:
http://www.python.org/workshops/2002-02/papers/05/index.htm
Paul D. Fernhout
> And I was curious what they used in the DARPA challenge, but
> initially, to me it looks like the software and hardware are all
> patented and non-free.
Glad you are looking into this.
I wrote on that topic around 2001 to the Markle Foundation. I actually set
foot in the first ALVAN when it was being built around 1986 at CMU, the
precursor to all these things. I spent about a year hanging around CMU back
then (around Red Whitaker's lab and Hans Moravec's lab). DARPA was funding
all that work on ALVAN (Autonomous Land Vehicle). Some related old research
people did then:
http://www-old.ri.cmu.edu/pubs/pub_4196.html#abstract
"This document describes the local map blackboard component of a mobile
robot system, the Autonomous Land Vehicle (ALV), under construction at CMU.
This system will utilize diverse sensors including cameras, sonar, and range
finders to navigate autonomously on roads; it is composed of several
processes performing these tasks and communicating via a Local Map Builder
process, which is the subject of this document."
More publications by one of the authors there:
http://www-old.ri.cmu.edu/people/person_299_pubs.html
What I wrote related to all that:
http://www.pdfernhout.net/on-funding-digital-public-works.html
"""
* what have funding policies in automotive intelligence wrought?
Consider again the self-driving cars mentioned earlier which now cruise some
streets in small numbers. The software "intelligence" doing the driving was
primarily developed by public money given to universities, which generally
own the copyrights and patents as the contractors. Obviously there are
related scientific publications, but in practice these fail to do justice to
the complexity of such systems. The truest physical representation of the
knowledge learned by such work is the codebase plus email discussions of it
(plus what developers carry in their heads).
We are about to see the emergence of companies licensing that publicly
funded software and selling modified versions of such software as
proprietary products. There will eventually be hundreds or thousands of paid
automotive software engineers working on such software no matter how it is
funded, because there will be great value in having such self-driving
vehicles given the result of America's horrendous urban planning policies
leaving the car as generally the most efficient means of transport in the
suburb. The question is, will the results of the work be open for inspection
and contribution by the public? Essentially, will those engineers and their
employers be "owners" of the software, or will they instead be "stewards" of
a larger free and open community development process?
Open source software is typically eventually of much higher quality
http://www.fsf.org/software/reliability.html
and reliability because more eyes look over the code for problems and more
voices contribute to adding innovative solutions. About 35,000 Americans are
killed every year in driving fatalities, and hundreds of thousands more are
seriously injured. Should the software that keeps people safe on roads, and
which has already been created primarily with public funds, not also be kept
under continuous public scrutiny?
Without concerted action, such software will likely be kept proprietary
because that will be more profitable sooner to the people who get in early,
and will fit into conventional expectations of business as usual. It will
likely end up being available for inspection and testing at best to a few
government employees under non-disclosure agreements. We are talking about
an entire publicly funded infrastructure about to disappear from the public
radar screen. There is something deeply wrong here.
And while it is true many planes like the 757 can fly themselves already for
most of their journey, and their software is probably mostly proprietary,
the software involved in driving is potentially far more complex as it
requires visual recognition of cues in a more complex environment full of
many more unpredictable agents operating on much faster timescales. Also,
automotive intelligence will touch all of our lives on a daily basis, where
as aircraft intelligence can be generally avoided in daily life.
Decisions on how this public intellectual property related to automotive
intelligence will be handled will affect the health and safety of every
American and later everyone in any developed country. Either way, the
automotive software engineers and their employers will do well financially
(for example, one might still buy a Volvo because their software engineers
are better and they do more thorough testing of configurations). But which
way will the public be better off:
* totally dependent on proprietary intelligences under the hoods of
their cars which they have no way of understanding, or instead
* with ways to verify what those intelligences do, understand how they
operate, and make contributions when they can so such automotive
intelligences serve humane purposes better?
If, for example, automotive intelligence was developed under some form of
copyleft license like the GNU General Public License,
http://www.fsf.org/copyleft/copyleft.html
then at least car owners or their "software mechanics" would be assured they
could have access to the software in source form to ensure safe operation.
What might be "street legal" in terms of software modifications might be a
different story -- in the same way people can't legally drive with a cracked
windshield or a broken headlight. For example, software changes might need
to first be proven safe in simulation before being provisionally "street
legal". But, the important thing is, foundations or government agencies
funding code development could insist on some form of free licensing terms
for automotive intelligence as a matter of public policy.
There are many other areas of human activities that the exponential growth
of technology will effect. Automotive intelligence is just one of them that
is here now and which I am familiar with from tangential interactions at
universities with people developing it. In enough time similar issues will
arise for the software behind household robotics or intelligent devices that
assist the elderly or handicapped. The IBOT wheelchair by Dean Kamen using
complex software to balance on two wheels is just the beginning of such devices.
http://iwsun4.infoworld.com/articles/op/xml/99/11/22/991122opmetcalfe.xml
Note the IBOT wheelchair was developed entirely with private funds it seems,
so the reasoning in this essay does not apply directly to it. Also, in
general Dean Kamen is a role model of a socially responsible for-profit
inventor. Still, the issue arises of whether "Johnson & Johnson" should be
funding such development, as was the case, as opposed to, say, the "Robert
Wood Johnson Foundation", as was not, given the public policy issue of
whether individuals should be continually dependent for personal needs on
proprietary software. In either case it would be worth it to pay billions
for such innovation, and the public will pay that in the end as a toll on
for such devices.
There is a real question here of how our society will proceed -- mainly
closed or mainly open. It is reflected in everything the non-profit world
does -- including the myths it lives by. The choice of myth can be made in
part by the funding policies set by foundations and government agencies. The
myth that funders may be living by is the scarcity economics myth. How does
that myth effect the digital public works funding cycle?
"""
Anyway, have you seen this movie? It's interesting on the non-city race:
http://www.pbs.org/wgbh/nova/darpa/
http://www.hulu.com/watch/23347/nova-the-great-robot-race
Is there a web site (wiki) about your work? Or about other stuff people are
doing for integration? You could start a Google (or wherever) code project
and go from there, building something that interacted with VDrift. Or, maybe
you could work with that team perhaps. Though I still like running Gran
Turismo because you'll be tempted to cheat and look at the VDrift code or
access data structures and such, that you just could not with one of those
other simulators. Still, VDrift sounds like it would be easier to get
started with. You could grab the bitmaps and start processing them, and
probably it would be easy to feed something back for steering.
Here is another part of the simulation idea. (I suggested this recently to
Alain Kornhauser, who I used to work for long ago, and runs an effort at
Princeton, but I think he ignored it). Once you get you code to work with
the simulated output, then start introducing more and more noise into the
images of all different sorts. Static. Blotches. Random lines. Whatever. In
some frames, not in others, persisting for a while, and so on. So, that will
challenge your recognizer system to get better.
Note: the winning Grand Challenge car used vision integrated with laser
ranging. It might be possible to make a simulated laser ranger to go with
that VDrift software. You could read the ZBuffer depth results if it uses
OpenGL perhaps. You might in general make other virtual sensors by modifying
VDrift. And, as with vision, you might want a setting where you introduce
noise. So, the compass lags, or even gets stuck some times. :-) The inertial
guidance system would loose wheel rotations sometimes. And so on. I think
you could spend quite a bit of time on this and do some impressive stuff.
And it would be stuff that *interested* a lot of people out there. Because
anybody with a GNU/Linux box could play with those add-ons. Whereas not many
people can mod their cars. And, out of that, you might collect together a
team of interested people to the point where you would have enough interest
and ability, maybe even in conjunction with a local "hackerspace" place to
mod real cars eventually to try out the software or real. And probably then
completely rewrite it for some unexpected thing. :-) But, it's a step by
step thing.
I'd also be curious to see how existing Grand Challenge software might work
with simulations. Would it be really accurate? Or would it get confused?
Heath Matlock
<pdfer...@kurtz-fernhout.com> wrote:
> Is there a web site (wiki) about your work? Or about other stuff people are
> doing for integration? You could start a Google (or wherever) code project
> and go from there, building something that interacted with VDrift.
There's a mailing list thread about the possibility of the work :-) If
I'm not too lazy, I may setup some type of Google Code, Gitorious,
Github, and Sourceforge project as well as throw up a .deb for others
to find the work more easily.
> Or, maybe you could work with that team perhaps.
Who? Maybe Elben Shira who created Kapal would be interested, I
emailed him, but haven't asked if he would be willing to help. He did
point me to some interesting sources which I will paste below. Aside
from just asking people, usually open source projects work by someone
discovering it and realizing it's their life's ambition to pour in
some dev time, from what I gather.
> Though I still like running Gran Turismo because you'll be tempted
> to cheat and look at the VDrift code or access data structures and
> such, that you just could not with one of those other simulators.
> Still, VDrift sounds like it would be easier to get started with. You could
> grab the bitmaps and start processing them, and probably it would
>be easy to feed something back for steering.
You know, VDrift is indeed nice, however, TORCS was the game I was looking for.
It's easy to hop in create your own A.I., there are quite a few nice
tutorials explaining how to do this:
http://www.berniw.org/torcs/robot/robot.html
http://www.wdbee.gotdns.org:8086/
http://sites.google.com/site/torcscreations/drivers
And I found http://usuarios.lycos.es/f1torcs/
From what I can tell, allows you to create various vehicles and their
physics, which might which potentially be useful if you wanted to
test it with an ATV or any other alternative vehicle. TORCS is quite
interesting, there are quite a few competitions taking place with
universities participating in them regularly. However, VDrift has the
better physics engine, so it might be best to write a decent robot for
TORCS and then attempt to port it to VDrift. Or maybe skip this all
together and use The Player Project.
http://playerstage.sourceforge.net/
Elben writes:
"Ah yes, the simulation for the autonomous vehicle. We used
player/stage. It's not the best, but it's the best free one out there.
...Unfortunately, my research with the autonomous vehicle is owned by
the university so I can't give out the source code. But I can tell you
that most of my work is derived from Maxim Likhachev's research. He
has created some of the best state-of-the-art path planning
algorithms. Check out his PhD thesis. This paper (Likhachev, et al) is
a good starting point for Anytime Dynamic A*, a state-of-the-art path
planner."
Paul, thanks for your ideas, those are definitely some interesting
thoughts to keep in mind.