Greedy Model

This release consists of a simple, yet complete, architecture. In this architecture, the learning model is integrated in the agent's behaviour, which performs online learning.
The agent combines the prior knowledge of the domain, used for obtaining plans, with the knowledge learnt from interacting with the environment. This knowledge is used for choosing the best subgoal in a greedy fashion, only taking into account the first 'step' of the whole plan needed to complete the level.
This architecture is purely deliberative (has no reactive part yet), since the environment is completely deterministic for now.