My mechatronics team was assigned to make a robot that could collect 5 bean-cans on a field and drop them into a hole. The entire robot had to fit within a 1 ft^3 volume, but could unfold after it was started. Although our robot was camera-shy for the competition (it ran fine between official trials, but failed at every official trial), we got it working nearly every time in 10 weeks. (See it here.)
The team included Yuxi Chen, Brian Finn, and Jasjeet Seehra.
We originally considered two strategies to move the cans, picking them up or pushing them, but ultimately decided to go with the simpler of the two and just push them. We chose to use line following to navigate the board.
In order to complete the design, our team used rapid design/prototyping iterations to test out ideas quickly. This lead to us having nearly a month to test out code and performance before the due date. We also tested each of the subsystems separately and then slowly added each together.
Our first prototype was nothing more than a foam-core sheet with holes punched out for wheels and two continuous-rotation high-torque servos. We found we had to add a lot of ballast to get the traction needed to push the cans.
We made several other versions of the prototypes, all out of foam-core. We used foam core because it was easy to work with, had enough rigidity for what we were trying to do, and allowed us to experiment with dimensions. These prototypes allowed us to get our electronics working properly before settling on a final chassis and body.
Our final body was built from laser-cut acrylic, aluminum sheets (for the gates), and cardboard and foam-core for non-essential parts. We continued to need to add a lot of ballast for the heavier versions. (Our final robot had nearly 15 lbs. of ballast to get the traction we needed!)
We tweaked quite a bit in the last week, and so our robot turned red (we ran out of black electrical tape.)
Overall, it was a really rewarding experience, especially for how hard we worked on it. Our line follower worked well, but we had a really hard time line counting, and our final circuit was very sensitive to minor changes in the board. More virtual modeling would have helped, and given more time, changes to the circuit would have been very helpful.