Applications of fuzzy logic in heuristic control have been highly successful, but which aspects of fuzzy logic are essential to its practical usefulness? This paper shows that an apparently reasonable version of fuzzy logic collapses mathematically to two-valued logic. Moreover, there are few if any published reports of expert systems in real-world use that reason about uncertainty using fuzzy logic. It appears that the limitations of fuzzy logic have not been detrimental in control applications because current fuzzy controllers are far simpler than other knowledge-based systems. In the future, the technical limitations of fuzzy logic can be expected to become important in practice, and work on fuzzy controllers will also encounter several problems of scale already known for other knowledge-based systems. 1

This paper introduces a simple yet effective method for using causal domain knowledge for learning to control dynamic systems. Elementary qualitative causal dependencies of the domain are exploited in order to dramatically speed up the learning of reliable control strategies from a simulation model of the system. The reliability of the obtained control strategies is strengthened as well. The effectiveness of the method has experimentally been studied at the problem of learning to balance a pole.

Reinforcement learning attracted increasing interest in recent years. While reinforcement learners proved successful for certain problems with comparably small system state spaces, they tend to have severe problems when the system state space is larger. This paper introduces a method for integrating domain knowledge into the process of Q-learning, in order to allow significantly faster learning and good scale-up behavior for larger state spaces. We present experimental results in the domain of a simulated pole-and-cart system indicating the applicability and usefulness of the approach. 1 Introduction The automatic or semi-automatic generation of controllers for dynamic systems from a model of the system is of great practical importance. For complexity reasons, classical control theory deals essentially with linear models of systems. This limitation is perceived as an important issue because many practical systems are nonlinear when described in a `natural way'. An example is the clas...