Adaptive coarse coding for neural network controllers

Examines a class of neuron based learning systems for dynamic control that rely on adaptive range coding of sensor inputs. Sensors are assumed to provide range coded vectors that coarsely describe the system state. These vectors are input to neuron-like processing elements. Output decisions generated by these `neurons´ in turn affect the system state, subsequently producing new inputs. Reinforcement signals from the environment are received at various intervals and evaluated. The range boundaries adapt to form a topological map relating sensor inputs to the control outputs. This is an improvement over current rigid rectilinear partitioning methods. Preliminary experiments show the promise of adapting `neural receptive fields´ when learning dynamical control. The observed performance with this method exceeds that of earlier approaches