[Dprglist] self driving trucks and the desert

[Karim Virani]

Try not to take this as putting words in Paul's mouth. My guess is Paul was
not trying to come up with a rigorously defensible analogy, but
accentuating a difference of constraints in the moment of that conversation.

In reality, we all place constraints on getting from here to there. A path
following constraint is just one example, but it's almost never the only
one. It's also not required. If you have constraints against bumping into
things instead of trying to plough through them, then you probably have
some sensors involved. If you have a path following constraint there there
are also sensors involved in determining how far off the path you have
travelled. It could be GPS outside, it could be odometry - could be a dozen
other kinds of sensors. The path could be described in different kinds of
coordinate systems. Paths may be defined with a different stickiness - how
tightly you have to cleave to them. There may be a different stickiness per
waypoint. A path can be described without waypoints - could be formulaic.

Adding path constraints may help with certain kinds of navigation problems,
could be unhelpful, or could be required. It'll almost never be sufficient.
Path planning can get you around known obstacles (a building), less likely
to encounter obstacles (lane keeping), but won't be helpful in avoiding
collisions with dynamic obstacles (vehicles, people, donkeys). And for
highly planned paths, you have the burden of maintaining that definition's
correspondence with changing reality.

Animals (including people) follow paths all the time. There are all sorts
of reasons to do so. There are also reasons to depart from a path. It'll be
a long time before robots can judge dynamically which is the best approach
in a given situation. Until then, they follow the rules we lay out. But
your robots are following a path even if they only do it emergently. I'd
like to see them emergently avoid driving through a freshly planted flower
bed.

Path planning is just something in our toolbox. Why not learn how to use
it? Figure out where it works best and where it doesn't? The more I
discover about it, it's not a single kind of tool, it's a spectrum.

-not Paul

On Sat, Oct 23, 2021 at 11:44 AM David P. Anderson via DPRGlist <
dprglist at lists.dprg.org> wrote:

> Hi Paul,
>
> I've been reflecting on your comments Tuesday evening about the self
> driving trucks needing lots of waypoints to stay on the road, vs. the
> robot demos shown that evening using more sparse waypoints.
>
> In particular the observation that the robots are operating essentially
> unconstrained, as in a desert, except for the need to avoid obstacles,
> while the trucks need to stay centered on the roadway.
>
> I'm not sure that is correct.  It seems that the long hallways that my
> robot had to navigate is the functional equivalent of the trucks'
> roads.  So the robot is not in fact operating as in a desert,
> unconstrained.  But must instead follow a fairly narrow and constrained
> path.
>
> Now if that is the case, why does it not use hundreds waypoints to
> accomplish that task?  Or more to the point, why the army trucks do?
>
> So I pondered this for a while and it occurred to me that in the robots'
> case, the waypoint navigation is not responsible for keeping the vehicle
> on the "road," i.e., centered in the hallway.   Rather that is the task
> of a separate group of sensors and behaviors.   But for the convoy, the
> navigation behavior is responsible for navigation in the global sense as
> Chris was describing, and also in the local sense of staying centered on
> the road.
>
> Now for the army trucks to do it the way the robots do would also
> require a second suite of sensors and behaviors to stay on the road ---
> not trivial --- while the method you describe can all be done with just
> GPS, no other sensors required.   (Though I assumed or maybe you told me
> there are forward looking radar to keep from running into the truck in
> front of you.)
>
> In any case, the desert vs. constrained roadway is perhaps not the best
> analogy for what the robots are doing.
>
> cheers!
>
> David
>
>
> _______________________________________________
> DPRGlist mailing list
> DPRGlist at lists.dprg.org
> http://lists.dprg.org/listinfo.cgi/dprglist-dprg.org
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.dprg.org/pipermail/dprglist-dprg.org/attachments/20211023/249d661a/attachment.html>