Task-oriented reinforcement learning for continuing task in dynamic environment

This paper presents task-oriented reinforcement learning, a modified approach of reinforcement-learning to simplify continuing dynamic problems in a more realistic and human-like way of thinking from the viewpoint of the tasks. In this learning method an agent takes as input the 'state of task' instead of 'state of environment' and chooses appropriate action to achieve the goal of the corresponding task. The proposed system learns from the viewpoint of tasks that enables the system to find and follow a precise policy in a continuing-dynamic environment and offers simple implementation for a multiple agents system.