|
[DPRG] where to go? (was Long Haul Waypoints)
Subject: [DPRG] where to go? (was Long Haul Waypoints)
From: David P. Anderson
dpa at io.isem.smu.edu
Date: Wed Sep 13 16:06:50 CDT 2006
Hi John,
I used a Garmin Etrex Vista-C. I used the "average" mode for 5 minutes
for each waypoint.
For your second question, I think the key is that the robot must be able
to survive on its own, so to speak, without GPS. When it can do that
robustly, then adding a low priority GPS layer into the subsumption
behavior is not difficult, and need not be that accurate.
The DARPA Grand Challenge is not really the best example of this because
all the waypoints were on the road (no offroad capability required) and
their waypoints were VERY closely spaced, especially around curves and
so forth.
For our robots, we'd like to have just a few waypoints, ideally only 2,
(where we are and where we want to be) and have the robot figure out
how to handle the terrain and obstacles and the pathway in between.
The robot I have been working on, jbot:
<http://www.geology.smu.edu/~dpa-www/robo/jbot>
uses inertial measurement and odometry for navigation, and sonar for
obstacle avoidance.
For example, here is a video of the robot navigating repeatedly
between two waypoints located at the opposite ends of a garden.
<http://www.geology.smu.edu/~dpa-www/robo/jbot/jbot_gardenx.mpg>
The waypoints "pull" the robot from one end of the garden to the
other, and the sonar reflections "push" the robot away from the
intervening obstacles. If the waypoints were chosen by less-
than-perfect GPS measurements it would not really effect that
navigation.
Hence, the GPS measurements don't have to be in the center of the
path, if the robot can stay on the path by itself. Sonar and
Infrared are at this time maybe more practical than digging the
obstacle information out of a video stream.
best
dpa
> Subject: [DPRG] Long Haul Waypoints
> From: John C. Abshier jabshier at kc.rr.com
> Date: Sun Sep 10 13:01:54 CDT 2006
>
> Those points match the descriptions and pictures much better. What GPS
> did you use? I have been wondering, how do you tell a robot where to
> go? At my home in Leavenworth, I cannot get a WAAS signal with a Garmin
> Etrex Vista. It will not keep one on a 2 lane road. There are other
> GPS's that are more accurate. Also for some GPS you can post process
> the data for better accuracy. But may of the DARPA Grand Challenge
> teams reports occasionally having large GPS error with high priced
> subscription differential GPSs. Even without errors, maps have
> inaccuracies. If there is a small river next to a railroad, next to a
> road, the map maker makes the river, road, and rail road wider than real
> life to make them visible on the map, then displaces them to provide
> visual separation. I think the eventual answer will be vision. You
> tell the robot generally where to go and it uses vision to avoid
> obstacles, stay on roads/trails, and to recognize the final target
> location. Of course, vision remains an open research problem that
> resists Moore's Law.
More information about the DPRG mailing list
|
