Learning to improve agent behaviours in GOAL

This paper investigates the issue of adaptability of behaviour in the context of agent-oriented programming. We focus on improving action selection in rule-based agent programming languages using a reinforcement learning mechanism under the hood. The novelty is that learning utilises the existing mental state representation of the agent, which means that (i) the programming model is unchanged and using learning within the program becomes straightforward, and (ii) adaptive behaviours can be combined with regular behaviours in a modular way. Overall, the key to effective programming in this setting is to balance between constraining behaviour using operational knowledge, and leaving flexibility to allow for ongoing adaptation. We illustrate this using different types of programs for solving the Blocks World problem