Results

Visualization:

Validation:

Tests of our analysis code showed that we retained a small amount of drift in our data. However, this was mostly negligible, being on a scale of a maximum of 0.14m drift over an entire data set.

The flight reconstructed from our data looked very accurate to the actual event as seen in the video, with a sharp increase in height at the beginning followed by a slow decline. We also calculated that the disc spun about 1.5 rotations, which also seemed consistent with the video.

When comparing the measured distance traveled by our Frisbee to the distance calculated by our data. The actual distance was 3.35m, our reconstruction returned a distance of 0.2m, which is, let's say, a bit off. Since the plot qualitatively looked fine, we believe there is some sort of scaling issue somewhere in our analysis.

Potential Sources of Error:

Some of the error in our data could likely be accounted for by a couple potential sources of error.

One potential issue is drift in the accelerometer. When analyzing our data we assumed that our accelerometer was perfectly centered on the axis of rotation of the Frisbee. Of course, this is impossible to do exactly, we just got as close as we could. Having the sensor offset slightly would have resulted in small amounts of centripetal acceleration being recorded. In our analysis we would have assumed that this was linear acceleration. Because of this, we would have added it to the actual linear acceleration resulting in a small drift.

A more general source of potential error would have just been accuracy issues with the IMU. The IMU we ordered was pretty cheap, and so was likely not as accurate as more expensive sensors. This was compounded by the fact that the rapid movement of the Frisbee required us to increase the maximum value the accelerometer and gyroscope record, further lowering accuracy.

Implications:

The audience for our project (beyond just us), is Frisbee manufacturers and ultimate players interested in the creation of a more integrated 'smart Frisbee' type of product. Our deliverable was fairly rough, but it showed potential for a more serious product.

Most of the components were small enough to be easily integrated into a Frisbee. The battery was our biggest and heaviest component, but it was greatly over-spec'ed to our needs. A product that was made from the start to house these components would not have the flaws in handling and aerodynamics that our Frisbee did.

Using a more specialized board rather than the unnecessarily powerful micro-controller we used could reduce cost or provide funds to use a nicer IMU.

Our analysis showed that the data you get from an IMU can be processed and used to return useful information about the flight of a particular Frisbee throw. While we did not get the chance to explore more concrete ways to 'judge' a throw, we don't doubt that it would be possible with the data we collected.