Assisting Mechanicats with Code

By Arjun

Task: Help Mechanicats, the other DISD team, debug their code

Competition is always stressful for everyone. There's so much that can go wrong, and when something does, it feels like all your hard work has gone to waste. We know first hand how it feels when something breaks. That's why we volunteered to help out Mechanicats when there was an announcement over the intercom that a team needed help with vision.

Mechanicats were having some trouble with their vision code. They told us they were able to sample correctly when they were on depot-side, but when they were on crater-side, they were unable to correctly identify the position of the gold mineral. We talked to them and helped them debug their code, and after a bit of testing, we were able to identify the problem for them.

The issue was that the TensorFlow detector they were using was returning null when it had already been called before because it expected client code to cache the objects it returned. This meant that when there were lots of minerals in the background, the detector would reuse the same objects to be more efficient. Mechanicats did not realize this, and thus since they were not caching the response of the detector, when they were on crater side they were unable to detect minerals.

After we fixed this issue for them with a few simple lines of code, we helped them exhaustively test it to ensure that it worked correctly. Mechanicats was extremely grateful for our help.