Learning motor programs

Now consider learning a motor program for a computer interface. A simple, classic example is the video game Breakout, which was developed by Atari in 1976. The player turns a knob, shown in Figure 10.2. This causes a line segment on the bottom of the screen to move horizontally. The Paddle contains a potentiometer that with calibration allows the knob orientation to be reliably estimated. The player sees the line segment positioned on the bottom of the screen and quickly associates the knob orientations. The learning process therefore involves taking information from visual perception and the proprioception signals from turning the knob and determining the sensorimotor relationships. Skilled players could quickly turn the knob so that they could move the line segment much more quickly than one could move a small tray back and forth in the real world. Thus, we already have an example where the virtual world version allows better performance than in reality.

Figure 10.3: (a) The Apple Macintosh mouse. (b) As a mouse moves across the table, the virtual finger on the screen moves correspondingly, but is rotated by 90 degrees and travels over longer distances.
...nter.ps,width=2.8truein} \\
(a) & & (b)

In the Breakout example, a one-dimensional mapping was learned between the knob orientation and the line segment position. Many alternative control schemes could be developed; however, they are likely to be more frustrating. If you find an emulator to try Breakout, it will most likely involve using keys on a keyboard to move the segment. In this case, the amount of time that a key is held down corresponds to the segment displacement. The segment velocity is set by the program, rather than the user. A reasonable alternative using modern hardware might be to move a finger back and forth over a touch screen while the segment appears directly above it. The finger would not be constrained enough due to extra DOFs and the rapid back and forth motions of the finger may lead to unnecessary fatigue, especially if the screen is large. Furthermore, there are conflicting goals in positioning the screen: Making it as visible as possible versus making it comfortable for rapid hand movement over a long period of time. In the case of the Paddle, the motion is accomplished by the fingers, which have high dexterity, while the forearm moves much less. The mapping provides an association between body movement and virtual object placement that achieves high accuracy, fast placement, and long-term comfort.

Figure 10.3 shows a more familiar example, which is the computer mouse. As the mouse is pushed around on a table, encoders determine the position, which is converted into a pointer position on the screen. The sensorimotor mapping seems a bit more complex than in the Breakout example. Young children seem to immediately learn how to use the mouse, whereas older adults require some practice. The 2D position of the mouse is mapped to a 2D position on the screen, with two fundamental distortions: 1) The screen is rotated 90 degrees in comparison to the table (horizontal to vertical motion. 2) The motion is scaled so that small physical motions produce larger screen motions. The advantages of the original Xerox Alto mouse were scientifically argued in [39] in terms of human skill learning and Fitts's law [78,192], which mathematically relates pointing task difficulty to the time required to reach targets.

For a final example, suppose that by pressing a key, the letter ``h'' is instantly placed on the screen in a familiar font. Our visual perception system recognizes the ``h'' as being equivalent to the version on paper. Thus, typing the key results in the perception of ``h''. This is quite a comfortable, fast, and powerful operation. The amount of learning required seems justified by the value of the output.

Steven M LaValle 2016-12-31