RECONSTRUCTING LANE CHANGE BEHAVIOR IN SELF-DRIVING VEHICLES USSING HIDDEN MARKOV MODELS AND LONG SHORT-TERM MEMORY (LSTM)

• 主要作者: Batrisyia Chalid, Sarah
• 格式: Final Project
• 語言: Indonesia
• 在線閱讀: https://digilib.itb.ac.id/gdl/view/77542
• 機構: Institut Teknologi Bandung

Technological advancements have facilitated the execution of daily activities for humans through the incorporation of smart systems, a notable example being autonomous vehicles, commonly known as self-driving vehicles. One of the decisions that must be made by a self-driving vehicle is whether or not the vehicle can change lanes on the highway at a certain speed. The autonomous nature of self-driving vehicles causes the need for more protection in the form of vehicle insurance for self-driving vehicles. Vehicle insurance serves to protect passengers from the risk of damage that may occur to the vehicle due to an accident. This study aims to predict the lane change behavior and lateral velocity of self-driving vehicles on highways. The dataset comprises Next-Generation Simulation (NGSIM) data obtained from the US-101 road segment, commonly referred to as the Hollywood Freeway. The methodologies employed in this study encompass the application of both the Hidden Markov Model and Long-Short Term Memory (LSTM) techniques. The metrics for predicting vehicles that remain in their lane show a precision of 0.97, a recall of 0.97, and an F1 score of 0.97. For vehicles that shift to the right lane, the precision is 0.96, recall is 0.94, and the F1 score is 0.95. Furthermore, for vehicles moving to the left lane, the precision stands at 0.96, recall at 1.00, and the F1 score at 0.98. In the context of lateral velocity prediction, the LSTM model demonstrates a notable decline in RMSE and converging to a distinct value. The convergence rate of RMSE exhibits variability across the analyzed Vehicle ID sets. These predictions serve as a foundational criterion for autonomous vehicles when deciding whether to change lanes on highways.