Human in the loop learning in Autonomous Driving
Title: Human in the loop learning in Autonomous Driving
DNr: Berzelius-2023-346
Project Type: LiU Berzelius
Principal Investigator: Alkis Sygkounas <>
Affiliation: Örebro universitet
Duration: 2023-12-12 – 2024-04-01
Classification: 10207


As autonomous driving technology advances, ensuring the safety and efficiency of these systems remains paramount.In this paper we introduce an online approach of combining double dueling deep Q-learning (DDDQN) with human-in-the-loop (HITL) learning to enhance the performance of autonomous driving systems. This combination harnesses the computational capabilities of deep reinforcement learning while leveraging the real-time insights of human feedback.The results suggest that, integrating human feedback leads to a notable performance enhancement in the DDDQN algorithm, surpassing the efficiency and outcomes of baseline models trained exclusively with reinforcement learning. Perfmorm x\% better in convergance, robustness and transfer learning or scalability. For that we will use microsoft AirSim, tensorflwo gpu and other minor (needed) libraries.