Action Selection using Reinforcement Learning

Success of adaptive autonomous agent is evaluated by how well it could perform the desired task in dynamic environment, i.e. selection of appropriate action even if circumstances and environment changes. Agent cannot select appropriate action in random as well as in sequence because the environment in which agent acts are dynamic in nature. Hence, action selection always remains a central question, especially in the field of adaptive autonomous agents that can function robustly and efficiently in complex and dynamic environment (Blumberg 1994, p.22). Now in this situation, reinforcement learning, getting feedback (negative or positive) of performed action has very important role in decision making of future action selection with the help of present experience. Hence, though there are few difficulties in implementation of concept of reinforcement learning, it is still the first choice in adaptive autonomous agent building. Especially in this essay I will describe how reinforcement learning helps agent for efficient action selection in dynamic environment and I will also highlight the importance of reinforcement learning in action selection as well as its limitations.

Download full pdf text