|
[DPRG] Cheap (ish) SBC runs linux
Subject: [DPRG] Cheap (ish) SBC runs linux
From: Chris Jang
christopher.jang at yahoo.com
Date: Sat Nov 10 13:14:11 CST 2007
>> Does anyone else have an inexpensive,
>> Linux-running development board
>> with lots of I/O capabilities?
>
> I have been playing with the NGW100 from Atmel.
> It is promising (and complete)
> and only ~ $73 from Mouser.
Dale,
When you first posted about a VIA C7 based micro-ATX
mainboard and then corrected yourself, saying that it
is not embeddable, I thought - perhaps you were right
the first time. Let me explain.
About two months ago, I saw Mark Curry's robot,
Intrepid, win the SRS Robo-magellan competition. It
uses a mini-ITX Pentium M on a heavily modified COTS
RC truck. I learned a lot from seeing this machine.
Sometimes, you think about the next machine you wish
to build. Then once you see it in real life (because
someone else has done it), you start thinking some
more.
The SRS Robo-magellan required touching traffic cones.
This necessitates a high resolution imaging system,
either with cameras or LIDAR. The environment also had
a lot of GPS signal interference due to occluded sky
from trees, buildings and a metal roof covering part
of the course. So by design, guidance based purely on
an IMU/GPS is not good enough. Terminal guidance is
required.
People have joked that I want to turn a robot into a
PC. Well, that's what Intrepid is. It's even running
MATLAB for signal processing. This may seem ridiculous
- I used to think so when I was in school - all of
those EEs doing signal processing using MATLAB. But
now I have seen MATLAB used in production (kind of)
and read about PeakStream's (before they were acquired
by Google) conscious emulation of a MATLAB like API.
I've concluded that the math and software problems are
so hard that robots need to have a development
environment that is close to a workstation. This has
already happened for game consoles as they are really
workstations locked into an embedded application.
So I think it is not crazy at all to use a SFF
mainboard on a robot. Yes, it will require an
expensive battery for power. But that's ok - it's just
like a laptop (and could be a laptop). Intrepid is an
example of this design direction.
The biggest thing I saw from Intrepid's performance is
that it succeeded because it had robust behavior when
things went wrong. It had distinct states for
different phases of the mission profile. So when
starting and then underway to a waypoint, it relied
the estimated odometry and GPS. Sonar was for obstacle
avoidance. What happened is that it made it's way to
the loiter area around the waypoint. What actually
happened is that GPS interference had shifted the
loiter area. It was short of the waypoint. So it kept
trundling around and around, back and forth. From
about five to ten meters away, the camera by chance
picked up the cone. So the robot stopped and slowly
crept up on the cone. After it touched the cone (the
crowd of observers were clapping), the robot proceeded
to drive the 100 meters or so back to the starting
point.
jBot is another robot with very robust behavior. In
terms of overall mechatronic engineering, reliability
and operator ergonomics, it is the most advanced
amateur autonomous ground robot I have seen. But it
would not have been able to complete the SRS course
due to the cone touching requirement combined with GPS
interference.
So... I don't think it is bad to have a top heavy
robot with what is effectively a PC on it. If it makes
the robot much smarter, then it can be an effective
design. I think too this can be good for everyone -
people with either a software or hardware background.
Generations change and the clear trend is towards
managed languages using interpreters and dynamic
compilation. With higher level languages, this
(hopefully) makes software easier to develop.
Chris
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
More information about the DPRG mailing list
|
