A Human in the Loop Based Robotic System by Using Soft Actor Critic with Discrete Actions

In this paper, the possibility of extending the Soft Actor Critic (SAC) with human control feedback is proposed and implemented for mobile robot navigation using Turtlebot3 Burger. The Human In The Loop (HITL) technique is integrated into the reinforcement learning algorithm for obstacle avoidance-based navigation. The algorithm is usually presented in the continuous action domain and for this work, the SAC with discrete action settings is proposed to suit the minimal discrete space. The human feedback assisting the robot with the discrete action settings has proved to be better in path planning and navigation when compared to the standard algorithms with the episode number.