Inference of takeover time budget for level 3 autonomous vehicles using triboelectric sensors and hybrid learning

With the rapid development of autonomous driving technology, Level 3 (L3) autonomous driving systems are seen as a crucial transition towards higher levels of automation. However, L3 autonomous driving technology faces two major challenges in the dynamic adjustment of the Takeover Time Budget (TOTB): the real-time acquisition of driver's non-driving behavior information and the calibration of the TOTB corresponding to these different non-driving behaviors. To address these challenges, this paper introduces the Takeover Recovery Time Inference (TORTI) System, integrating digital gloves (D-Gloves) with hybrid learning. D-Gloves, featuring triboelectric sensors, can accurately track the driver's hand movements and interactions with various objects in the car, thereby converting the driver's non-driving behavior information into electrical signals in real time. We establish an End-to-End correlation between driver's non-driving behavior data and Takeover Recovery Time (TORT) through hybrid unsupervised and supervised learning methods, achieving 90.3% accuracy. Furthermore, we conducted a series of comparative takeover experiments to validate the dynamic adjustment scheme for TOTB based on the TORTI System. Our research not only inspires new ideas for human–machine interaction but also showcases its potential value in the application and implementation of L3 autonomous driving vehicles.