[ExI] Fwd: Cars and Ants

[Emlyn]

2009/5/21 Mike Dougherty <msd001 at gmail.com>:
> On Wed, May 20, 2009 at 9:28 PM, Emlyn <emlynoregan at gmail.com> wrote:
>> But I like my second idea better, which relates to ants. I'm thinking
>> of a pheromone trail. Could we get cars to leave a weak chemical trail
>> as they drive around? If most of the cars did this, you'd end up with
>> a nice aggregate trail in every lane of every road, strongest in the
>> average path, with a nice falling gradient as you move away from the
>> center. So it should be quite useful to follow it.
>
> Have you considered using gps enhanced with cell-tower triangulation?

I have! I think we'd use that extensively, of course. Can we get the
millisecond and perhaps millimeter precision that we need, though,
with that? Also, what do you do when there are dropouts? I think any
system totally reliant on remote infrastructure is going to have these
bad failure modes that render the thing unusable.

However, I think it's so close to usable, that you don't need much to
fill in the gaps. The pheromone idea is one idea for filling in the
gaps.

> That would cover most of the already developed infrastructure in metro
> areas.  If we still aren't sure about the accuracy of road maps, we
> could allow the cars to train on the roads while under human control.
> This would free us from the daily commute once we've driven to work
> once or twice.

Yep, training modes are a great idea I think. If the early systems
only work for routes you frequently travel, where the car has trained,
that'd be fine. Most of us do our most boring and repetitive
travelling daily, that's the biggest bang for the buck.

> If the cars on the road were 802.11n+ ad-hoc network
> routers, we would be able to query conditions ahead as far as there is
> traffic density to carry the network.  This would also allow each node
> in the network (car) to alert position information to nearest
> neighbors to not only increase collision avoidance but also inform
> intention for the sake of changing lanes or entering/exiting the
> roadway.  (ex: I need to exit right soon, but I'm in the leftmost
> lane.  With intention broadcast, neighboring cars can momentarily open
> a path)

Also you could potentially swap route training info, and you could
probably send coordinate info around too; if you can pinpoint the
location of another car in relation to yourself, and it can tell you
where it thinks it is, then there's something to check your own idea
about where you are against. Networks of cars are a great idea.

This is a very bottom-up approach, which is not a way we are used to
thinking about the roads I think. With respect to things like
signalling intentions to change lanes, you'll get all kinds of
incentives to cheat; there's no contract between you and other cars
for them to behave in the way you want them to.

A reputation system would probably help. Something like what's
embedded in bittorrent. Maintain measures of  reputation, and decide
to cooperate based on that. Iterated tit-for-tat.

>  The open relay of car-to-car data is very similar to your ant
> analogy, but uses RF rather than chemical signals.  Considering the
> seemingly ever-present stream of cars at nearly every bend in the
> roads that I drive, I wouldn't need greater persistence than real-time
> (or point-in-time) traffic data.  Once the cars are equipped with
> sufficient technology, then we can discuss enhancing hundreds of
> thousands of miles of roads.  An "old car" is 10+ years old.  The
> average age of highway is usually several times that of the cars on
> it.  An entire generation of cars will be replaced before a typical
> road is refurbished.

That's what I liked about the pheromone idea, it doesn't require road
refurbishment. It mightn't even require new cars!

Dunno how it holds up in rain.

>
> I too have thought about this a great deal.  :)

Awesome!

-- 
Emlyn

http://emlyntech.wordpress.com - coding related
http://point7.wordpress.com - ranting
http://emlynoregan.com - main site