Rain-Aware Lane Change Decision Model For Autonomous Vehicles Using Deep Reinforcement Learning

  • Ahmed Alzubaidi

Student thesis: Master's Thesis


The autonomous vehicles have shown great potential in the past few years, where there is an evident increase interest on their adoption in our roads. In order to have fully reliable autonomous vehicles, they must cope with adverse weather. One type of adverse weather is the presence of rain on roads which causes serious safety risks to the road users. Autonomous Vehicles requires a Lane Change Decision(LCD) model which instructs the AV to change lanes as needed, which is a crucial aspect of the decision making of the AV. The literature includes a number of attempts of utilizing deep reinforcement learning techniques to propose an LCD. However, none of the existing solutions have considered the presence of rain in the development of the LCD, which is precisely the gap found in the literature, this work attempt to address. This work has proposed rain-aware LCD, utilizing deep reinforcement learning. Specifically, this study has exploited Deep Q-Networks (DQN). For the simulation task, an open-source simulation tool was leveraged, known as Highway-env. The simulation tool had to be modified to simulate road behavior under specific rainfall conditions. To evaluate the proposed solution, we compared it with a rule-based LCD, known as MOBIL, in terms of efficiency, safety, frequency of lane changes, and the global impact on the road. The results obtained allowed us to conclude, that our solution guarantees higher safety with achieving almost similar performance in regard to efficiency. Furthermore, it was demonstrated that our solution obtains identical or similar global impact when compared to MOBIL.
Date of AwardJul 2022
Original languageAmerican English


  • Autonomous Vehicles
  • Lane Changes
  • Deep Reinforcement Learning
  • Adverse Weather.

Cite this