(Somewhat Flexible) The AI agents developed must utilize Fuzzy logic in some way. Fuzzy Inference Systems, Fuzzy Trees, NeuroFuzzy architectures, etc. are all allowed and encouraged. Other architectures and methods may be allowed on a case-by-case basis but require approval from the organizers. This rule exists to encourage usage and application of Fuzzy in relevant (and fun!) problems.

Any machine learning method (or none) is allowed including reinforcement learning, evolutionary/genetic algorithms, and expert knowledge encoding.

Teams must maintain their code in a repository and utilize typical repository management best practices. The repository may be public or private, but must be shared with the organizers before the agent deliverable due date in order for organizers to access the correct version of the code.

The solutions must be able to interface with the game in Python as described in the Controller class in Kessler, but there are no other prescriptions for development style or coding paradigms. E.g. one team may decide to code in pure Python while another may compile some heavy calculations associated with their controller using C++, Cython, mypyc, etc. but then have a Python wrapper to interface with Kessler. Both of these are allowed with the stipulation that all competition teams must resolve all potential compilation issues for target hardware with the organizers well before the final agent delivery date.