View Full Version : Fit-Bot - a Fit-PC based rover

10-27-2008, 10:48 AM
Greetings all,

I am new to the board and fairly new to building robots. I wanted to build a rover that I could use to experiment with autonomous operation, telepresence, etc. I already have a Lynxmotion A4WD1 rover for the platform, with the Sabertooth 2x10 motor controller and a Phidgets servo controller. My original plan was to mount a small laptop without the lcd display as the brains, but I recently discovered the Fit-PC Slim (http://www.fit-pc.com/new/fit-pc-slim-specifications.html), a very small PC based on the Geode LX800 cpu that can run either Linux or Win XP, has built-in wireless and usb, and requires only 6 watts and can run on a 12v battery. It looks to be about the size of a pico-itx, not as much cpu power, but should be sufficient for my purposes.

The Fit will leave plenty of room on the rover for some sensors and a webcam. I already have a DLink DCS-5300G (http://www.dlink.com/products/?pid=342) but may get a Logitech Quickcam Pro 9000 (http://www.logitech.com/index.cfm/webcam_communications/webcams/devices/3056&cl=US,EN) and mount on a servo pan/tilt kit, also controlled with the Phidgets controller.

I will be coding in VB6 because I am too old/lazy to learn a new language, so I will be searching the download area for example code. So thanks in advance for any help. I will post up pics when I get everything together.

10-27-2008, 12:04 PM
I look forward to your pics coppertubing.

10-27-2008, 12:55 PM
It depends a lot on what you consider telepresense. Manhandling 640x480 video on a stripped-core 800Mhz machine just doesn't happen in reality. If limited resolution or low quality video is good enough, then it can happen reasonably - just don't expect it to be doing anything else (running other threads at the same priority).

Certainly you shouldn't expect to be doing image processing and image recognition on something that slow/stripped except for *very* low resolution and *very* limited application.

Welcome to the forum, btw!

10-27-2008, 01:29 PM
Yes, I realize I won't be doing any image processing on this thing. I'll be happy at first to have a remote controlled mobile webcam.

10-27-2008, 02:03 PM
A lot of simple image processing can be done at very low resolution however. I run 160x120 resolution on J5 when I'm doing object tracking for the smoothest operation, doesn't affect the quality of the tracking at all, in fact improves it.

10-27-2008, 02:41 PM
You know...its odd. I wonder. The ER-1 ran on 500 mhz pcs with 500 megs of ram, could be run over the internet, (slowly), and had incredible object recognition and obstacle avoidance. I'm not sure how they did it, but they did...I don't know exactly where I'm going with this, but it seams like more can be done with less with an efficient system.


10-27-2008, 03:03 PM
Because they had hardware accelerated image processing, DB. That hardware was constructed for them for that specific task. Never underestimate the power of coprocessing. ;)

Tybs - depends what you're doing. Simply tracking colors - yes, that's a low-resolution job. Identifying complex shapes, or working on almost anything outside of color from a distance - that's a high-resolution job.

If you're looking at a triangle that fills the entire frame, you have 19,200 pixels to work with there. If you have a triangle that fills a quarter of the frame, you have 2400 pixels to look at. If it fills 1/16 of the frame you have only 600 pixels. So the wider the view and the more you move the object away, you very quickly run into not being able to define the edges or shape of an object. And that's actually why your color tracking works better at lower resolution. Less definition to confuse things. ;)

Another thing to consider: Not all Mhz are created equal. The Pico has parallelization of instruction - the Geode doesn't. An identically clocked Pico doing graphics should be at least a dozen times faster.

I have a few different Geode-based boards here that I was looking at for video decompression for settop boxes. None of them, even the multi-processor boards, could handle NTSC video anywhere near realtime.

Copper - acknowledging that you'll probably be growing out of it in the immediate future, sounds like it may fit the bill for you. I'm not trying to be a downer, just don't want to see anyone buy something expecting more than it can reasonably deliver.

10-27-2008, 03:10 PM
what hardware? it was a kritter cam plugged into a usb port. I'm not sure about the stepper motor controller that they used, but the video stuff was all on the PC. (i'm pretty sure)

11-19-2008, 12:49 AM
The ER1 does all of its video processing on the PC, there is no hardware assistance derived from the RCM. The RCM controls the motors and can provide access to an array of analog and digital inputs/outputs even though Evolution never really supported it. My latest iteration of the ER1 is/was based on an Intel mini-itx mobo with the dual core Atom processor and the object recognition is just amazing. I recently burned up my RCM and can't find any schematics for it so it is a dead bot right now but when it was running it was fantastic.


10-06-2009, 08:32 AM
I look forward to your pics coppertubing.

Finally after almost a year, pics are up in this thread (http://forums.trossenrobotics.com/showthread.php?p=34923#post34923).