Optimal Path Finding using Iterative SARSA

This paper presents a novel and state-of-the-art algorithm named Iterative SARSA to effectively determine the optimal trajectory for an autonomous mobile robot present in an unknown environment. Additionally, a detailed comparative analysis of the proposed algorithm is provided along with other traditional reinforcement learning algorithms using apropos parameters such as path length, computational time, and execution risk. The traditional algorithms used here are Q-learning and SARSA (State-Action-Reward-State-Action). Based on the calculation of next step, these algorithms use either of two reinforcement learning methods - the on-policy or off-policy. While SARSA and Iterative SARSA use the on-policy method, Q-learning utilizes the off-policy method. Optimized trajectory planning along with obstacle avoidance has always been a challenging yet foundational component of various principal applications. Being one of the most primary algorithms of machine learning, any development using Iterative SARSA should render a greater applicative scope.