Vision Logic Open File Report

Chapter 4: Single Dimension Scanning Vision - 2

Updated 9/21/09

 For our next step, we have the robot moving and talking while using its 9 pixel horizontal image scanner for navigation using vision alone. Here I'll show you some small movies (mpeg format), that illustrate the visual capabilities - however limited, that can be achieved with a very low resolution visual setup. Sometimes the robot is hard to understand, and I'll key you in on what it is saying on each movie clip.

Basic Concept

 In the arena, when the robot is far from the black target, it sees with its 9 pixel vision after histogram clipping, this - a 180 degree field of white walls with a distant black target as one pixel wide. The range here is about 2 feet.

 When the robot is much closer, say less than six inches, this is what it sees - a much larger black target filling its field of view.

 The point to be made here is that if you merely count the number in each frame of black continuous pixels, you can tell how close you are to the target. When the robot is far away, you may only record one black pixel. But when close the number will fill up a maximum of 6 or 7 since a flat target cant be picked up on the 0 and 180 degree views.

Movie 1
 Well start with a bright light source, and move it in front of the robot at different angles. The robot sees the light, and responds by reading out the angle of the light source. (If its the same as last time, it says nothing). the robot will say at each new position "Target at - (three digit degrees). For example, left is zero degrees and right is 180. The position read is the center of each 9 pixel bin.
 To calculate the angle, the robot takes the white bits from the histogram sliced image, and does a centroid calculation. Here it finds the center of the mass of white pixels and calculates the center. This is the angle reported.

Movie 2
 Putting this into action, the robot drives directly toward the light, first orienting its body toward the light, then making corrections at periodic intervals to re-aim. Scanning is very slow and time consuming! but you get the idea. Robot is saying "Target Dead Ahead"...
 After the robot makes a centroid calculation on the image, it measures the angle, and knows how much to rotate to point at it by on simple calibration I made - I found the time to make the robot turn 180 degrees with the motors on and the robot calculates the partial angles and thus time to rotate to the targets measured position.

Movie 3
 This is a very important step here. The last white light demonstration the robot drives toward a non point source, calculates its center and moves toward that centroid. THEN when the lamp is at a specific distance which equates to the number of white pixels in its view, it stops just before it hits. This is a huge step in demonstrating the advantage of a multi pixel vision over just a two pixel light seeker. It knows how close it is to the lamp, and can aim right at the center no matter how big the light is. You cant do THAT with a two pixel eye!
 In this instance above, the robot will stop when it sees the glowing globe fill about 4 pixels. This is about 2 inches.

Movie 4
 Now here's a new feat for our vision guided robot, to drive to a BLACK target of finite width, and stop when it is close without hitting it. No glowing lamps this time.
 The centroid calculation is the same, but we now are concerned with dark pixels. The white walls read as 1's and the black target will read as a series of 0's in the robots sliced histogram vision. Again, when the robot reads a total of 4 or more continuous black pixels, it assumes it is at the target. If we push the robot right against the target, it reads about 7 pixels...

Movie 5
 Here I've added more information on what the robot reports. As the robot nears the target, It reads out the width of the target each time. You can see when it is far, it will read 1 pixel wide, but when close that value goes up until it stops just before hitting. The robot is saying "Target width 1 pixels" or what ever width is sees.

Movie 6
 Finally, the icing on the cake so to speak, after driving to the target, the robot docks with the black foam board and stops, similar to what it might do if connecting to a battery charger. But here we dock with 9 pixel vision ONLY and don't use IR beacons to guide us in!
 The robot reports its range with pixel widths, and when it is front of the target, it says "I am in front of the black target", and "Now docking with target" and finally after docking, which is determined here by impact with the front central bumper plate, it proclaims "I am docked with black target".


For this series of experiments, we were able to demonstrate some of the capabilities of having a visual field of 180 degrees and a very low resolution of only 9 pixels wide. In a Biomimetic sense, this shows that increasing the number of photo sensors with some crude directionality from one or two light spots as in simple organisms to what can be best described as primitive arthropod vision of wide angle, and one dimensional will have a big advantage. It can be seen here with the robots similar visual acuity, that our primitive arthropod would have had more information about its environment and had an advantage in escaping predators by being able to not only determine which side the dark cave was for cover, but at what approximate angle. The animal could have then rotated directly toward the cave and made its rapid escape. The poor hapless worm with only one or two eye spots would be slow to find cover and become food for the predator.

You can see the evolutionary push here - More pixels in your eyes, the better you can survive. This trend of course continued and after hundreds of millions of years of evolution, trilobites had thousands of facets in their crude eyes. The trend continues today. Dragonflies have the most "pixels" of any insect, about 750,000 per eye. You can imaging the changes in their tiny brains to be able to process all of that. Blows my mind...

In our home robotics realm, the advantages of being able to seek and dock with black or white targets is a big advantage as well, and this additional sensory improvement may make the difference between running out of charge and shutting down and seeing the docking port.

In our next experiments, we will be having the robot count targets in its vision, size them up and perhaps select the larger to drive to. Beyond that, a whole series of avoid maneuvers using 9 pixel scanning vision while moving is planned.