Autonomous Vehicle Steering Based on Evaluative Feedback by Reinforcement Learning

Steering an autonomous vehicle requires the permanent adaptation of behavior in relation to the various situations the vehicle is in. This paper describes a research which implements such adaptation and optimization based on Reinforcement Learning (RL) which in detail purely learns from evaluative feedback in contrast to instructive feedback. Convergence of the learning process has been achieved at various experimental results revealing the impact of the different RL parameters. While using RL for autonomous steering is in itself already a novelty, additional attention has been given to new proposals for post-processing and interpreting the experimental data.