Vision
To test the reliability of our ingredient localization, we developed a stress-testing dataset. This dataset contains two configurations, a staircase ingredient layout, and a pseudo-random layout. Additionally, we varied the lighting conditions, including adding shadows to the scene. The results of this stress testing are shown in Table 1.
| Ingredient | Staircase Patter Success Rate (%) | Pseudo-Random Success Rate (%) |
| Meat | 98 | 98 |
| Cheese | 100 | 96 |
Manipulation
After our initial manipulation subsystem implementation, we performed testing on its ability to successfully pickup and place various ingredients. This served two purposes. First, it provided metrics on the reliability of the manipulation subsystem, and second, it doubled as a trade study of various ingredients. For the testing, we performed 30 ingredient pickups tests per ingredient and logged both if any slices were picked up and placed, and if so, the number of slices that were manipulated. It is important to note that this testing was done with the old, low-flow suction system, and our eventual reliability with the higher flow system is much higher.
| Ingredient | Pickup Success Rate (%) | Slices Per Pickup |
| Mozzarella | 100 | 1.1 |
| Cheddar | 56 | 1.21 |
| Provolone | 33.33 | 1.2 |
| Bologna | 73.33 | 1.45 |
| Honey Ham | 46.67 | 1.07 |
| Turkey | 0 | N/A |
| Genoa Salami | 60 | 1.17 |