Date: Wed, 7 Jun 2006 01:12:04 -0500 (CDT)
Subject: steering strategy
X-UID: 176
Content-Type: APPLICATION/octet-stream; name="design20060606.png"

Adding in robot control support for the remote control is pretty easy. 
There are only three new commands. These are in addition to the existing 
manual driving through the gamepad. Manual override is full-time even in 
autonomous mode. That's kind of a principle I've learned from having 
things go wrong before.

(applies to speed strategy only)
SPEED - set the upper and lower motor duty cycle limits

(applies to speed and steering strategies)
PARK  - set speed to zero and try to center front wheels
DRIVE - heartbeat that allows the robot to drive, without this, it not go

It's disappointing to see all of the bang-bang control and conditional 
logic for steering. The speed strategy will be similar. It does not feel 
sophisticated in that I have math guilt (should be using more). But many 
abstract ideas disappear when faced with coding something real and making 
it work on limited hardware.

The speed strategy and road segmentation must be coded next. For this 
pass, I'm not going to do anything sophisticated. I'll just use histogram 
intensity thresholding with a very simple road model. This works when the 
light is right. I want to at least see the robot drive by itself on a path 
once and validate the basic approach.

Last week, I realized that I've been making this problem more complicated 
than it must be. All the robot needs to know is:

1. Do I turn left, right, or go straight?
2. Do I speed up or slow down?

(Note, I'm completely ignoring reverse - the robot only drives forwards.)

The goal is to answer these two questions in any way possible. There is no 
necessary requirement for any particular internal world representation or 
solution. It may be that tricky algorithm X works well. But it may also be 
that something much simpler exhibits equivalent performance in realistic 
conditions. So instead of presuming a complex solution, I'm starting from 
the assumption that something simple is adequate.

I know that I've bounced around from extremes of philosophy on this 
project. Right now, I'm back to a very practical, hands on, let's see it 
work mode of thinking. As best as I can discern, without practical 
experience or formal training, there is a tendency to wild preconceived 
assumptions based on very little evidence. Then once there is more 
progress and experimental results, reality takes hold and shatters the 
preconceptions. That's experience.