Personalized Curriculum Design Improves Autonomous Vehicle Risk Management

The article discusses the importance of tailoring autonomous driving (AV) training to individual drivers’ needs and abilities through a curriculum-based approach called CLIC. The authors compare the performance of CLIC with an untrained AV model and demonstrate its superiority in managing complex scenarios while maintaining proficiency in simpler cases. They also present qualitative results using SUMO GUI to visualize experimental results and support their conclusion.
The authors highlight that CLIC prioritizes more important or informative transitions for learning, which leads to better performance in challenging situations. In contrast, the untrained AV model has shown poor performance in complex scenarios despite good scene understanding. The article emphasizes the significance of individualized curricula in AV training to ensure transferability and practical usefulness.