Date: Mon, 20 Mar 2006 02:15:02 -0600 (CST)
Subject: working efficiently
X-UID: 152

I slept most of Sunday which is too much. It was actually convenient 
though because the power went out for about five hours due to the 
thunderstorms. I know exactly when as I now have a UPS that beeps at me 
whenever line power is lost.

I just hacked in 900 MHz radio support for the remote control robot 
driving software. It worked the first time. This is good! It should give 
sufficient line of sight range and greater reliability. I've found that 
WiFi tends to be unstable outdoors, even at short range.

Turning resistance of the front wheels increases sharply when the robot is 
stationary. It is enough to reach the stall point of the gearmotor. When 
the robot is in motion, the resistance drops so is not a problem. As the 
human operator, I'm aware of this and so compensate by effectively rocking 
the robot to free up the wheels.

The current system does not implement proportional steering. Left and 
right movement of the stick spins the gearmotor at full duty cycle in 
those directions. There's no direct linkage between stick and wheel 
position.

So what should be done is proportional steering when the robot is moving 
and non-proportional steering when the robot is stationary. I'll have to 
do this later today.

One big disadvantage of nighttime driving is that it doesn't make for good 
video. It's just too dark to see anything. So as the short term goal is a 
good demo, there's really no point to driving at night. I'm going back to 
focusing on daytime driving. This means the cameras must be able to pick 
out the road in images.

I'm hoping that at low bit rates (high compression), the discrete cosine 
transforms involved in JPEG remove all of the high frequency components in 
the image for me. Then region segmentation can be done based on color. 
This is all pretty dicey to hack together quickly.

I did go to the DPRG contest at the warehouse this last Saturday. They had 
three contests for indoor robots. One thing that was very noticeable is 
that about half of the robots lacked any feedback from the environment. 
They relied purely on wheel odometry. This meant that they were unstable. 
Completion of a course relied rather much on chance. Even very low 
frequency feedback corrections would have allowed many of them to correct 
for drift adequately well.

In response to the near rolling of my robot, one person said that SUVs 
roll all the time. This made me think that the robot really is a fairly 
good analogue to a miniaturized DARPA Grand Challenge vehicle. It handles 
like a SUV with a high center of gravity.