Determination of Reinforcement Learning Reward Parameters to Solve Path Planning of Unknown Environments by Design of Experiments

Abstract

The Reinforcement Learning Approach (RL) is used to solve the path-planning problem of an autonomous mobile robot in unknown environments. Despite that RL is a recent and powerful tool, it requires a lot of training processes because there are so many parameters in the agent’s training process. Some of these parameters have a larger effect on the convergence of the learning process than others, so, knowing these parameters and their suitable values makes the training process more efficient, saves time, and consequently makes the trained agent execute the required task successfully. No analytical equations are available to determine the best values for these parameters, therefore, in this paper, a statistical analysis is made using the design and analysis of experiment (DoE) methods to determine the parameters that have the largest effect on the training process. After that, analysis is done to determine the values of the most effective parameters. Results show that the determined parameters lead to a successful autonomous path planning in different unknown environments

Keywords

• autonomous path planning
• design of experiment
• mobile robot
• reinforcement learning
• reward
• training parameters