Genetic and reinforcement-based rule extraction for regulator control

This paper proposes a novel system for rule extraction of regulator control problems. The system employs a hybrid genetic search and reinforcement learning. The learning strategy requires no supervision and no reference model. The extracted rules are weighted microrules with a discrete nature that constitute a rule-based/table look-up structure capturing control actions. As an example of what the proposed algorithm can learn. The authors chose the problem of the trailer truck backer-upper. The system is capable of extracting rules that back up the trailer truck from arbitrary initial positions and show improved performance compared to a neural network controller trained with backpropagation through time