back-based control rather than open-loop control, I wrote an additional chapter of the course notes for the Robo-Cup students that explained the differences and trade-offs between open-loop and feedback-based control. I also discussed the need for calibration of sensor values to local conditions.
A number of students became fixated on using feedback control in surprising ways, however, spending a considerable portion of development time creating a feedback controller to make their robot able to drive in a straight line. Here is how one student explained his system:
The robot is now able to drive in a straight line thanks to its optical shaft encoders. The software continually samples both left and right sensors and adjusts the exact power level to the motors to compensate for any fluctuations in the motion of the robot. The algorithm itself is a combination of differential analysis and Newton's method, along with an "adjustable window" of correction. Simply stated, analyzing differentials gives a reasonable algorithm for adjusting the power levels of the motors, and the adjustable window minimizes excessive wobbling. This algorithm gives us as much accuracy as the resolution of the shaft encoders permits, while keeping wobbling reasonably low. ( Martin, 1994, p. 143)
The ability to drive perfectly straight, however, often does little to help a robot in its overall ability to solve the contest task. Although the activity of driving straight is based on negative feedback control, a strategy that employs this ability in a central way should be considered open-loop at the next higher level, as the robot would be performing a task with little feedback from the crucial environmental features! Therefore, these students' belief that they were using the preferable feedback control was misguided.
Taken as a whole, the collection of biases revealed in students' thinking suggests important misconceptions and misunderstandings about what are effective ways to build reliable real-world systems.
The examples in this chapter illustrate that students who participate in the robot design course have a variety of preconceptions about systems and control. These ideas are formed by experiences in the traditional academic curriculum and warrant examination specifically because they are not particularly effective when applied to the robot design task.
In the section entitled Robotic Control, we saw how many students have trouble seeing the contest task from a robot-centric perspective, and instead imagine the task from their own omniscient perspective. This naiveté is tempered when they try to put their ideas into being and realize that the robot's point of view is very different from their own.