Shown at left is the robots
      latest appearance. The wheel encoders version 1 is shown.
 

      
 Home brew
      wheel encoders.

      
My first attempt
      at making wheel encoders on each wheel was a success. But I discovered
      more problems with both the wheels, encoder and method than I
      bargained for! The plan was to put a rotating rim on the inside
      of each tire, and count the holes. I made the rim by carefully
      marking and punching out the holes, every 10 degrees so that
      the wheel had 36 positions. The optical interupter was mounted
      on a bracket - seen here as the black rectangular object just
      above the encoder wheel. It actually works mechanically very
      well, and I was able to count the holes very accurately to put
      the robot exactly at the distance I wanted and make accurate
      turns.

      
But while that
      worked ok, the original wheel - a hard rubber one, worked only
      on perfect smooth surfaces. In a real world home environment,
      the wooden parquet tile floor - the wheels failed miserably and
      slipped on ever bump or crack. It even got stuck on a micro piece
      of sand.



      
Next, I went back
      to the original foam wheels you see here. They grip the floor
      great, and walk right over any imperfections or sand/lint. Testing
      showed that the robot could still not go straight even with the
      encoders counting exactly the same for more than five feet or
      so. The problem is the differential traction on the floors for
      each wheel still made it up to six inches off the line after
      six to 10 feet. Also, turns were giving me a whole plethora of
      problems. While I can accuratetly set up the right wheel counts,
      the bigger the angle turned, the less accurate it was from wheel
      slippage while rotating. Even worse, different floor surfaces
      gave hugely different results! So I would program a 90 degree
      turn, and on a smooth floor it might come close. But on the wood,
      I got much less, and on carpet... totally up predictable.

      
Finally, dust really
      sticks to the wheels and disk. While turning, the static charge
      increases wildly, and it picks up sand, dust and whatever and
      deposits it directly on the photo sensor. Geez. So I can now
      come to the following conclusions from these extensive experiments:

      
1. The wheel encoder
      must be sealed and no fine dust can enter.

      
2. Only one wheel
      is needed to measure the distance traveled. Both do not add any
      accuracy because the robot can count the same number on both
      wheels and its still wont be on the line.

      
3. We must live
      with the fact that the turn angle will be somewhat dependent
      on the surface.

      
4. To follow a
      straight course toward a goal, the encoders on the wheels are
      not sufficient, and additional navigation methods will have to
      be employed over 3 to 6 feet or so. 

      
Update 12/27/10

      
Update 11/25/10


HOME