A BEHAVIOR-BASED ARCHITECTURE FOR REINFORCEMENT LEARNING

摘要

PURPOSE: the construction reinforcement learning of Behavior-based control, which is provided to, guarantees enough research high-performance without by action selection strategy. ;CONSTITUTION: the strategy of the agent architecture learning behavior of Behavior-based control is by the adaptation inquiry learning of self reinforcement to be modified to dynamic context. Behavior basic body architecture includes sensor, actuator, world model, external behavior, internal model, interior behavior, state supervisor, rewards calculator, evaluates estimator and an interpreter. The value that the state supervisor calculates Boolean state variables is defined the environment that support reinforcement learning determines current state. ;The 2011 of copyright KIPO submissions