This blog post is part of Makers Jönköping entry to Pi Wars 2024. Pi Wars is an international, challenge-based robotics competition in which teams build Raspberry Pi-controlled robots and then compete in various non-destructive challenges to earn points. The competition takes place over a single weekend and at the end of each day, prizes, medals and trophies are awarded to those competing.
As always we like to stay at the forefront of technology and enjoy the glow of the state of the art. So we wanted to use this new AI technology everybody is talking about – just because, because why not, right?
Well to be more specific, we wanted to train a machine learning model to recognize barrels for the eco-disaster challenge and use the model for navigating the eco-disaster arena.
This ambitious idea of employing machine learning to solve the problem would also include training the model from a simulated environment.
After pain-stakingly recording images of our 3d-printed barrels and annotating them, we trained the model with some promising results.
Further training yielded even better results (which is what I try to say to my children every day to no avail). Even Australians will be pleased to no longer be classified as Green Barrel.
