Servo - December 2011

FIGURE 6. A nice little package of 18 Dynamixel MX- 28 motors from Robotis.

are useful for specific control strategies. It is possible to make the motor operate stronger in one direction which is useful for a hexapod leg where it is mostly applying a force in a single direction. If set right, a person can physically move part of the robot without causing much strain on the motors. However, the simple linear or proportional feedback ( P ) nature of the onboard controller results in position errors or overshooting, and does not follow any standard control system strategy. If a derivative feedback term ( D ) were to be added, then the adjustable slope could be increased to help tighten up the error and increase the speed of the position adjustment. To polish things off, an integral feedback term ( I ) could be implemented to get rid of the error entirely. The motors use a high quality potentiometer that enables continuous rotation. The potentiometers work by having a resistive material in a ring with a metal brushed wiper attached to the shaft. Long days of repeated motion in the same place causes the resistive material to wear. This is highly unusual, and I am inclined to believe that other factors such as hot temperatures helped induce wear. The resulting wear causes a huge misreading for feedback. The drastically changed value results in a violent shaking behavior of the motor. The only way to remedy the problem is to replace the potentiometer, which is not too bad using a screwdriver and soldering iron.

4. Software.

The more I write code, the better I become. Considering that the code currently operating the hexapod was written about two years ago, my skills have vastly improved. The current user interface for the demonstration mode with the Sixaxis controller is very limited. It is difficult to see what is going on when the fit-PC2 is cranking away at processing face tracking and kinematics algorithms. Also, the use of the Sixaxis controller is seamless but limited. The face tracking algorithm was only able to effectively work at 11 frames per second at a down sampled image resolution of 160x120. This was the most effective I was able to have the hexapod track a face. The operation is not very robust as a person must be facing right at the camera with no tilt of their head in any direction. The face tracking mode is very difficult for most people to interact with.