Inverse Reinforcement Learning via Deep Gaussian Process for Reactive Collision Avoidance in Dynamic Environments

This paper deals with an automatic vehicle which navigates from the start to goal point avoiding the moving as well as static obstacles in dynamic environments. This research study proposes a novel approach for reactive collision avoidance based on Inverse Reinforcement Learning (IRL) which can learn complex reward structures with few demonstrations based on Deep Gaussian Process (DPG) to inherit the features from the normal distribution of the data collected from the LIDAR sensor. The simulation results showed that the accuracy of the proposed approach was higher than that of other approaches.