What I am hoping to see is the Petra AI code stubbed out for ML/RL tasks, https://github.com/0ad/0ad/tree/master/binaries/data/mods/pu... ... nearly every manager/function there could have it's actions modified based on what it views or calculates in previous gameplay. One was introduced a few alphas back before the interface, https://github.com/agentx-cgn/Hannibal