Driving Style Recognition at First Impression for Online Trajectory Prediction

TL;DR

The paper addresses online trajectory prediction for surrounding vehicles under limited observation by introducing driving style as a predictive cue. It proposes a hybrid offline-online framework: offline learning uses 13 car-following features with PCA to obtain two principal components and K-means to form three driving-style clusters, for which IDM parameters $\Theta=(v^*,T,d_{min},a_m,b_{comf})$ are estimated; online, two fast recognition methods—distance-to-cluster-centers in PCA space and local Maximum-Likelihood calibration with prototype parameter sets—are used to select style and predict the next trajectory over a $5\mathrm{s}$ horizon via the IDM. The experiments on the I80-1 dataset show RMSE reductions up to $37.7\%$ versus literature parameters and $24.4\%$ versus not performing style recognition, with method 2 generally outperforming method 1, including very short observation durations. The work demonstrates that first-impression driving-style recognition can significantly improve short-horizon trajectory predictions while keeping computation manageable, and it lays groundwork for integrating additional data sources such as roadside sensors.

Abstract

This paper proposes a new driving style recognition approach that allows autonomous vehicles (AVs) to perform trajectory predictions for surrounding vehicles with minimal data. Toward that end, we use a hybrid of offline and online methods in the proposed approach. We first learn typical driving styles with PCA and K-means algorithms in the offline part. After that, local Maximum-Likelihood techniques are used to perform online driving style recognition. We benchmarked our method on a real driving dataset against other methods in terms of the RMSE value of the predicted trajectory and the observed trajectory over a 5s duration. The proposed approach can reduce trajectory prediction error by up to 37.7\% compared to using the parameters from other literature and up to 24.4\% compared to not performing driving style recognition.

Figures (7)

• Figure 1: Framework of the proposed approach for driving style recognition and trajectory prediction.
• Figure 2: The relationship between SSE value and the number of clusters
• Figure 3: Driving styles clustering results. The method for driving style labeling ("relatively aggressive", "neutral," and "timid") is explained in the next subsection.
• Figure 4: The calculation method of RMSE for a single vehicle over a 5s duration
• Figure 5: The relationship between the prediction errors (RMSE) and data available ($t_{dur}$) for online driving style recognition method 1