Adaptive modular reinforcement learning

The adaptive modular reinforcement learning system was proposed to apply the reinforcement learning into more realistic control problems. It is composed of some control modules and a selection module. Its all modules are calculated by using the incremental normalized Gaussian networks (INGnet). It learned the task, where the “AND” condition of two types of sensor information should be discriminated, more quickly than the previous modular reinforcement learning, whose modules were calculated by using CMAS, or the reinforcement learning using INGnet. Since the number of the processing unites of the adaptive modular reinforcement learning was smaller than that of the modular reinforcement learning using CMAC or the reinforcement learning using INGnet, it is considered to have the ability to make more appropriate representations for the control.