GeoBot
      1


      
Sonar
      Array Layout and Technique


      
Updated 3/18/06

      
A crucial portion
      of the outdoor navigation of the robot will be a quad matrix
      sonar array, which I will go into some detail of it here. For
      daylight outdoor navigation and obstacle avoidance the usual
      light sensing type arrays will not work in the sunlight. You
      wont be able to use IR proximity or ranging, lasers are very
      hard to see unless they are very powerful and heavy, and vision
      based systems using CCD cameras will not be reliable in fields
      of rocks that are all exactly the same color on a cloudy day
      with no shadows. 

      
The first part
      of a solution to outdoor navigation for our robot was to install
      extremely durable high impact instrumented bumpers on the front
      and sides. The second phase of protection and direction seeking
      will in in the form of an array of 4 sonars projecting a specific
      pattern interpreted by a dedicated PIC processor as to not load
      the main controller. Here is a photo tour of our method, and
      some diagrams and schematics to follow on.

      

 The
      best configuration for a mapping sonar system such as this will
      be as close to a point source as possible as to ensure proper
      overlap of scan sectors and a bare minimum of dead zones. Our technique works like
      this: for sonars with approximately 30 degree patterns (we measured
      about 25d) are mounted in a small as possible array on the front
      center of the robot. Each in turn sends out a sonar burst and
      receives its return before the next one in line is fired. The
      whole scan takes about 1/4 second. The sonars beam is divided 

 into
      four zones according to the distance measured. 0 - 12 inches
      for example are the first zone out, and the second  at 12"
      - 24". The third zone is at 24" - 36", and finally
      anything beyond that we really don't care about for our purposes
      and is classified as the general zone 0. Using this zonal approach has many advantages
      and can generate complex movements amongst obstacles with relatively
      simplistic programming. The greatest dangers lie closest to the
      array, and thus have priority over the more distant obstacles
      and thus we use priority arbitration to control the robots response
      to the range of zonal dangers.
 
 So
      here is how it works. Three rocks which are too big to drive
      over are within the sonar arrays range as seen here. But which
      one to act upon? Will it be the huge rock to the right but far
      away, or the close small one which is about to strike the robot
      and in its path? The
      sonar processor uses priority arbitration architecture to ferret
      this out, and reports via serial or parallel lines to the main
      processor 


the
      obstacle of  GREATEST danger to the moving robot.
      In this case, the small rock is the danger, and zone 8 will be
      reported to the main controller.
 
 Enough
      of theory, lets make this thing. Rather than a hemispherical
      shaped housing to hold the sonars which was way to big to be
      mounted on the bumper, a very CAD optimized shape was created
      out of lexan as seen here. The sonar barrels are mounted vertically,
      spaced at exact 30 degree spans, and were offset both in X and
      Radius to create the compact housing shape seen here. This housing
      will have to take rock impacts because its on the bumper, so
      must be very durable!
 
 Here
      is the final scale drawing of the array from the top. the entire
      4 part array is just over an inch deep, and only 4 inches wide.


 Machining
      the housing per drawings, we found it made an attractive yet
      functional bumper addition. The wires bundle goes through a hole
      in the bumper, and routes up about 16 inches later to the sonar
      processor through flexible tubing.
 
 Top
      view of array on bumper.


 First
      bench testing on a flat surface with REAL rocks revealed a critical
      balance between detecting the ground as a reflection and an actual
      rock. The angle of the array was adjusted vertically as to pick
      out a 2 inch rock at all positions to avoid dangers. The ground
      reflections on gravel made for about a 2 - 4 foot reading.
 
 The
      sonar chip is the one to the lower left of the main 16F877a processor.
      I've hooked the LCD display to it directly to be able to see
      whats being seen by the 4 sonars. There are two rows of four numbers on the
      display. The top row is sonars 1 - 4 actual inches display from
      the nearest reflection. The bottom row is the same data, but
      displays the zone number of the corresponding reflection. These are the numbers in
      which the Priority Arbitration architecture will deal with.


Out in
      the field, here is a sample reading as it is moving toward the
      rock pile. I've enlarged the display and superimposed it on the
      robot for clarity! Obstacles
      range from a cozy 30 inches out, to 18 inches. The choice here
      would be to veer left, to avoid the two closest obstacles. Choices
      for the different zones will be different. For the closest zone
      (all sonars) we would have to stop, then look for the clearest
      path out. For the second zone, we bank right or left while moving,
      the third zone means to veer right or left gradually using the
      PWM differentially. And finally, zone 0 means keep going.
 
Here
      is the schematic for the sonar processor. You can of course click
      on this one to make it much larger. 


HOME

FastCounter by bCentral