Interaction-Aware Trajectory Planning with Multi-Modal Predictions for Unprotected Left Turns

In complex unprotected left-turn scenarios, autonomous vehicles (AVs) face trajectory planning challenges due to time-varying interactions and potential conflicts with multiple surrounding human-driven vehicles (HVs). However, existing methods often struggle to identify truly high-risk HVs while accounting for their uncertain driving intentions and behaviors, which limits planning safety, smoothness, and realtime performance. To this end, we propose an interaction-aware trajectory planning method. First, we establish an interactive behavior model that incorporates multiple human driving features. Then, an entropy-based approach is utilized to quantify interaction intensity and identify interactive HVs (IHVs). Finally, we construct a future scenario set and generate the trajectory via an interactive multi-scenario planning algorithm based on the Model Predictive Path Integral (MPPI). We evaluate the proposed method using the INTERACTION dataset and the CommonRoad simulation platform. The results demonstrate that our method achieves a well-balanced performance across efficiency, comfort, and real-time capability compared with other methods.