Ball Balance

Working from home has proven to have its challenges. One of which is that my home desk was not as level as my office desk. Regrettably, this lead to my thinking ball consistently falling of the desk and general productivity was reduced. To remedy this the logical solution was not to remove the ball from the table or level the desk. Instead a ball balancing platform was the clear solution to this problem and the best use of my time.

Core concepts:

The objective of this project was to implement a PID controller to balance a ball. The discrepancy of where the ball is relative to the target position is used as a feedback to the PID controller. This platform implements a simple two degree of freedom (DOF) platform to adjust the roll and pitch in order to keep the ball at a target location.

For this application a resistive touch screen was used to determine the location of a ball on the platform. A resistive touch screen consists of two layers that are separated by an array of very small spacer dots to prevent the bottom rigid layer from contact the flexible upper layer when no pressure is applied to the screen. When pressure is applied to the top screen the spacer dots compress and allow for a contact to be made between the top and bottom layer of the screen. Both the bottom and top layers have a small resistive coating on them. This allows electricity to pass along the X-Axis on the top layer and along the Y-Axis on the bottom layer. Essentially you are creating two perpendicular resistors. By perpendicular I am referring the direction of the current across the top and bottom layers of the screen. The ability to sense the touch location is based on applying a voltage across one axis at a time and applying a high impedance voltage read on the other axis. We will use the case there the x-axis has a voltage across the top layer. Because the y-axis (the read axis) is in a high impedance mode no current flows along the read axis. Thus, by ohms law, there is no voltage drop on the sensing layer because there is no current over the resistor. The microcontroller is constantly switching between read axes to derive the x and y touch location using a simple voltage divider calculation.

Image derived from: https://www.sparkfun.com/datasheets/LCD/HOW%20DOES%20IT%20WORK.pdf

Room for Improvement:

There is a positive feedback associated with tilting the platform when attempting to affect the balls movement. This platform has a fixed fulcrum, thus this fulcrum is probabilistically never projected through the center of the ball. When the fulcrum is not projected through the center of the ball, a pitch or roll creates a counter productive centrifugal force on the ball. One solution to this is to increase the DOFs of the platform. Having additional DOF would allow for “softly catching” the ball. By this I mean, if you know where the ball is going you can forward project where the platform should be to stop the ball motion by increasing the normal force on the platform and minimize the counter productive centrifugal forces. This would be accomplished by ensuring that the effective fulcrum is at or projected past the current location. Projecting “past” would be relative to the ball movement and the desired response of the ball. By this I am referring to the fact that you would want your fulcrum projections to be projected on different sides of the ball based on if you wanted to encourage or discourage the current motion.
The project currently does not linearize the servo motor angles. The PID controller is directly tied to the servo requested angles and thus is attempting to use a linear solution to a nonlinear problem. This approach holds up for small angle change request as the linear movement associated with 1 degree is approximately the same as 2 degrees. As expected, this becomes less reliable as the PID requested response becomes significantly different than the steady state, for example when a step response is requested.
Having thin wood substrate and hobby grade linkage the platform has very loose tolerance. The tolerance stack associated with platform flex, linkage flex and touch screen resolution has proven to be insufficient at producing the sub-micrometer accuracy required. With that said, now the ball only falls of the desk when I am not around to catch it. So I would call this project a success.

Alternative Implementation:

I would have really liked to develop a solution that involved a capacitive touch screen. Actually, it was more about the fact that I wanted to use a tablet or computer monitor. Unfortunately, all nonindustrial touch screens that I am aware of use capacitive technology for touch registration. Using a tablet or computer touch screen would have opened up many fun and interesting opportunities. I spend some time researching a solution for creating a ball that would emulate the capacitive touch necessary to register a touch. I was unable to find anything practical for my application. The industrial application screen was out of the price range I wanted to spend on this project and would involve a lot more development friction. Alternatively, I could have overlaid a resistive touch screen on a non touch screen. The biggest issue with this was a difference in aspect rations.

The most cost effective and least compromising solution was to combine a computer monitor with a vision system to get the desired effect and benefit I was after. I could still do gesture recognition however it would have to be implemented with vision processing instead of touch registration. With that said, this actually allowed for greater flexibility in development. Unfortunately, for this part of the project I do not have much more documentation than the video below. This video demonstrates that stability of the target location even with a high percent of occlusion. This project also performed very well because of the high level of control over lighting. Because the camera was saturated from the monitor output it pretty much did not matter what the lighting conditions where. Additionally, the backlighting from the monitor caused an eclipsing effect that allowed the ball targeting to be very stable.