RecOpt: A Recommender-Based Optimization Framework for Enhanced Multi-Objective Decision-Making in Autonomous Driving

Autonomous driving systems operate in highly complex environments, necessitating real-time, balanced decisionmaking across multiple objectives such as safety, efficiency, and passenger comfort. This study introduces RecOpt, a novel framework that combines recommender algorithms with advanced optimization techniques to enhance multi-objective optimization (MOO) in decision-making. RecOpt tackles the challenges of large-scale mixed-integer MOO by identifying and restructuring key decision variables, thereby reducing the optimization scale and increasing efficiency. By integrating recall and finesorting technologies inherent in recommender systems, RecOpt efficiently screens and selects optimal solutions, surpassing the speed and scalability limitations of traditional MOO methods. Experiments utilizing the NGSIM dataset validate RecOpt’s efficacy in balancing safety, comfort, and efficiency, minimizing decision-making latency, and adapting to dynamic environments. Furthermore, RecOpt offers data-driven insights into explainable decision-making, providing a robust and transparent model for autonomous vehicle decision processes.