These last two weeks focused heavily on testing and system integration, as well as parameter tuning and output display. I helped Caroline with ultrasonic sensor integration and helped debug the code for reversing back from a wall, as well as turning away from an obstacle. I also worked on system debugging when we ran into an issue of short-circuiting due to lack of insulation underneath the Arduino board.
We also tried including a third scent for propane spray after discontinuing testing with smoke (it was unfeasible to be collecting smoke data indoors with extremely flammable substances in the vicinity) and collected data for the same. However, our sensors did not produce features that would be useful in identifying propane/butane. We decided to stick to classification with ambient, paint thinner and alcohol scents. After neopixel integration, I worked on running testing trials with our two scents and collected data for performance metrics: classification accuracy, time to convergence, distance to first positive scan, number of obstacle run ins etc. We did testing in the 1300xx hallway as well as TechSpark for the upcoming demos. I also helped make presentation slides for the Final Presentation and Poster.
The next two weeks will focus on more testing with our final tuned algorithm.