Driver Digital Twin for Online Prediction of Personalized Lane-Change Behavior

Connected and automated vehicles (CAVs) are supposed to share the road with human-driven (HDVs) in a foreseeable future. Therefore, considering mixed traffic environment is more pragmatic, as well-planned operation of CAVs may be interrupted by HDVs. In circumstance that human behaviors have significant impacts, need understand HDV make safe actions. this study, we develop driver digital twin (DDT) for online prediction personalized lane-change behavior, allowing predict surrounding vehicles’ help technology. DDT deployed on vehicle-edge–cloud architecture, where cloud server models behavior each based historical naturalistic driving data, while edge processes real-time data from his/her maneuver. The proposed system first evaluated human-in-the-loop co-simulation platform, then field implementation three passenger along an on/off ramp segment connecting through 4G/LTE cellular network. intention can recognized 6 s average before vehicle crosses lane separation line, Mean Euclidean Distance between predicted trajectory GPS ground truth 1.03 m within 4-s window. Compared general model, using model improve accuracy 27.8%. demonstration video watched at https://youtu.be/5cbsabgIOdM .