This week the team put in a lot of effort to integrate our software and hardware into a working system. I wrote the rest of the logic for the software pipeline, including obstacle avoidance and rotation upon not detecting bottles. Serena and I figured out how to command the Arduino from the Jetson to spin the intake. We all worked to mount components such as battery packs, Jetson, Arduino, and Realsense on the structure using tape and velcro. We tested the full integration by running the software pipeline on the fully mobile robot. We were able to solve many of the problems and bugs we ran into, such as the angle calculation overshooting when we put the structure on the iRobot when it worked fine when we tested on just the iRobot by itself. We also did initial testing with our 1st test benchmark, which is 3 fixed bottles within a 1.5 meter radius. We aren’t getting the accuracy or speed to reach our use case metrics yet, but will work to improve them next week.
Overall, I believe we made great progress this week and are on schedule. Next week, I have a couple ideas to try to improve the accuracy and speed of our system:
- Do angle calculation inside multitracking instead of angle calculation separately- this will improve speed and accuracy, however it will make it harder to detect obstacles that were originally out of view
- After the first inference that detects the bottle, use a different, faster object detection library to find other objects instead of Yolov5 which takes a whole second
0 Comments