Humanoid Robot

Another major initiative I've been involved in is the improvement of some aspects of the lab's anthropomorphic robot, which was largely constructed by previous experimenters. To be useful for experiments, the robot needs to be able to precisely position its two cameras along their pan and tilt axes. Later experiments will require precise movement of the three-DOF neck joints as well.

Thus far my work has centered around these areas:

Camera Pan/Tilt Units

The two "eyes" of the robot are held by pan/tilt units, which were designed by previous experimenters to be lightweight versions of commercial pan-tilt units used in previous experiments. However, all of the units showed several degrees of backlash. I traced this problem to runout in both the worms and worm gears, caused by excessive tolerances in the machinist's attempt to modify the worms to match the shafts to which they were mounted. To fix this, I purchased new gears and precisely turned the shafts to match ID of worms (light press fit), reducing the backlash to less than 1 arc-minute, sufficient for the experiments.

Fixation Algorithm

The robot is currently controlled by National Instruments' hardware and software. High-speed video acquisition hardware receives camera images, and the motors are controlled by stepper controllers. To test the dynamic performance of the improved pan/tilt units, I developed a LabView program to fixate the robot's cameras on a single LED held by an experimenter. Although simple (proportional gain plus damping), the algorithm was sufficient to show that that the robot can move smoothly and track objects.

Robot Neck

As part of a previous project, undergraduate Mech Eng students had designed and constructed a three-DOF neck to allow the robot's head to be tilted in pan, roll and yaw directions. Unfortunately, the students neglected to include gravity in their calculations of required motor torque, causing them to specify grossly-underpowered stepper motors to drive the neck joints. Further, they used coarse-pitch spur gears which added significant backlash.

To allow experiments to proceed as quickly as possible, the decision was made to replace the stepper motors with low-backlash high-torque servo motors. While the servos' 0.3 precision is not as good as the stepper arrangement used elsewhere, they were very easy to integrate with some modifications to the existing neck components. In the picture below, the original stepper motor is shown at left, and the servo at right.

In future, the neck will likely be redesigned completely to provide more accuracy and rigidity. Work on the robot is currently suspended while I finish the High Speed Display Panel.



Back to Projects