Date: Sun, 5 Jun 2005 18:20:35 -0500 (CDT)
Subject: robot outdoors part 1
X-UID: 57
Content-Type: IMAGE/JPEG; NAME="img1597.jpg"
Content-Type: IMAGE/JPEG; NAME="img1601.jpg"
Content-Type: IMAGE/JPEG; NAME="img1602.jpg"
Content-Type: IMAGE/JPEG; NAME="img1603.jpg"
Content-Type: IMAGE/JPEG; NAME="img1604.jpg"

I'm still debating different steering servo configurations. While there is no 
one right way, there are definitely wrong ones. Unfortunately, I do not know 
what those are as I lack experience. Mechanical design is like everything else 
in engineering, software included. When you have no experience, you are 
basically guessing.

Unpowered with the rear wheels decoupled from the drive shafts, the robot does 
roll straight. There is no vibration or tendency to track left or right. The 
rear elastic has enough tension to pull the wheels into position.

The nylon cord at the front of the vehicle is a placeholder until I figure out 
how to arrange the steering servo (which I still must make). Servos are 
typically referred to as closed loop control. That means that the servo control 
electronics actively hunts for the commanded position. But from another point 
of view, a conventional servo is open loop. Upstream, you have no idea what the 
true actuator position is. If the servo should become stuck, the electronics 
will catch on fire or shut down when the current increases above a threshold.

My intent is to use a linear slide potentiometer connected to cables. The 
microcontroller will use this in a voltage divider to know where the gear motor 
mechanism is. It's unconventional as most solutions use a rotary potentiometer.

Another thing I noticed is that the coupling between the rear drive axles and 
the wheels is a failure point. A lot of stress is concentrated there. If a 
coupling should fail, then the corresponding motor will have no resistance and 
instantly spin to high rpm. The other side will still be turning the wheel. The 
result will be the robot will want to spin.

I have no plans for monitoring current to each drive motor or placing encoders 
on the wheels. So the only way the robot will know a rear coupling has failed 
is when the gyro indicates it is turning when it should not be.

Yes, there could be a lot more instrumentation and fault monitoring 
electronics. Or the robot could be built better. But even with all of that, the 
problem doesn't go away. The robot needs internal fault detection. There are so 
many things that could potentially fail that the robot needs to have some self 
awareness of its own integrity for performing the mission, like a modern car 
(especially something like a Prius) or a late generation Tomahawk missile 
(which can self destruct if it believes it can not complete the mission 
according to parameters).