Unfortunately, I was unable to achieve accurate and reliable localization results using my ToF sensor data and the provided localization program, which I documented in my Lab 11 writeup. Even though my polar plot of the ToF sensor readings looked like the robot's surroundings, the localization algorithm would predict the wrong location the majority of the time. If I had more time, I would continue to debug my localization since it is the most reliable way to complete Lab 12; however, I had to leave campus early, so I will not be using localization in my map navigation, as it will likely provide innacurate results.

I implemented my navigation algorithm using code that resembles a state machine, where the robot can either be in a "move forward" state or a "rotate" state. I defined a function, angle_turn(), which uses yaw feedback to turn the car a specified number of degrees. It calculates the difference between the current yaw and begin_angle, which is the yaw at which the robot began rotation, and stops when the desired number of degrees has been reached. There are a total of 13 states, one for each movement the robot takes to traverse the map. After each movement is completed, the state_change_flag is raised; this flag is handled at the end of the code, where the begin_angle variable is updated to the current yaw, the state variable is incremented to move on to the next state, and the state_change_flag is reset to 0. All of this code is nested within the while (central.connected()) block, so it begins execution when the robot connects to my laptop over Bluetooth and continues to run in a loop until disconnected.

Here is a video of my robot navigating through the maze and successfully hitting all of the waypoint squares. Throughout the testing process, I observed that the performance of the robot varied drastically depending on the voltage of the battery supplying power to the motors, and this caused more inconsistency than I had anticipated. The rotation of the robot at each waypoint was not affected much because it was close-loop, but the forward movements often fell short or overshot the desired distance. Since the entire navigation sequence is 13 states long, and the success of hitting each state depended on the accuracy of the state before it, it took a while to tune each movement to be accurate; even in the successful run video, the robot is not perfectly centered at every waypoint. However, I tried my best within the limited time that I had, and I hope the result is sufficient to show that my robot is able to perform navigation using my semi-closed-loop algorithm.

I worked on this final lab with Zin Tun. During the last few labs, her robot has sometimes had trouble with connecting via BLE. In Lab 12, both my laptop and her laptop were unable to connect to the Artemis board onboard her robot; after attempting many methods to try and debug this issue and consulting many TAs during open lab hours, we decided to combine our efforts onto my robot for Lab 12. Thus, Zin and I worked together to incrementally build and test our Lab 12 code. Our process was to work on each state one at a time, adding a new state whenever the previous states had been tuned sufficiently. Even though the battery problem caused some inconsistency with our results, we were able to have a few good runs in the end to complete our final lab.