[extropy-chat] Pentagon-Sponsored Robot Race Ends As All EntriesBreak Down

Eugen Leitl eugen at leitl.org
Mon Mar 15 09:58:43 UTC 2004


On Sun, Mar 14, 2004 at 09:33:10PM -0800, Spike wrote:

> When you are driving somewhere, what is your mind doing?

You have no idea what your mind is doing, if all you 
use is introspection. 

> You may not be aware of the thoughts regarding your
> driving, but clearly there is a great deal of processing
> going on.  So what is your mind's driving related top
> priority?  

If you look at what retina's doing, it doesn't seem to prioritize a lot. It's
mostly hardwired. The later stages build models, but we don't know how. 
 
> I would argue that your mind's top driving-related
> priority is not guiding the car to your destination.
> That is priority number four.  Priority one is avoiding
> the smiting pedestrians.  Priority two is avoiding smiting
> solid objects.  Priority three is keeping one's Detroit 
> upon the pavement, the black part, not the lighter colored
> concrete that contains the afore-mentioned pedestrians
> and solid objects.  Priority four is navigating
> toward your destination.  I *hope* your brain's priority
> is in that order.

I can guarantee you that your visual system doesn't have priorities you can
write down cleanly. It does seem to latch upon moving objects, and edges, in
that priority. In locomotion navigation angle optical flows seem a good
source of information.
 
> In the next few weeks I hope to organize a team to enter
> the (probably) 2006 DARPA challenge.  Your assignment,
> if you wish to contribute ideas, is to think about
> what your brain is doing when you drive.  I leave

This would seem as a major sidetrack. I can guarantee you thinking about what
you do when you're driving will give you irrelevant if not actually
misleading information. If anything, stick a person driving a racing
simulator into an fMRI setup. That's way too coarse, but it will give you
lots more info.

> you with this one observation:
> 
> The DARPA Grand Challenge course had many mechanically
> sophisticated entries, none of which succeeded, yet the
> course was run successfully by eight vehicles, all with
> human drivers, all mass produced.  They were ordinary
> unmodified Dodge pickup trucks, which were to be used
> as chase vehicles carrying DoD officials holding kill 
> switches to stop wayward robots.  
> 
> This is a huge clue: the winning vehicle need not be 
> mechanically sophisticated.  The simplest quarter-ton 
> pickup can easily go faster than state-of-the-art driving 
> software can guide a vehicle.  We saw all sorts of sophisticated
> desert racing gear, nearly all of it unnecessary and undesireable.

No one knew that in advance. The original requirements asked for an
aggressive push for speed, in a rough terrain, possibly littered with traps. 
 
> I will propose a pickup truck with a small sensor package
> consisting of GPS, radar or lidar, sonar, compass and clock.
> Sufficiently sophisticated software can figure out what
> to do with inputs from those instruments only.

These requirements are a) entirely ad hoc b) unnecessary harsh. Some
signals contain more easily extractable information than the others. It is
not possible to tell that in advance, without having field experience.

I notice you're omitting visual input, while people drive almost entirely by
visual cues. You're also dropping inertial navigation, GPS by itself is
insufficiently precise. I would also use high-resolution maps of the area,
including custom maps.

That assumes we're sticking to commercially available sensorics, I'm sure a
lidar with wavefront timing could yield more easily extractable info than an
optical flow box.

-- Eugen* Leitl <a href="http://leitl.org">leitl</a>
______________________________________________________________
ICBM: 48.07078, 11.61144            http://www.leitl.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE
http://moleculardevices.org         http://nanomachines.net
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20040315/7bb81bd5/attachment.bin>


More information about the extropy-chat mailing list