An Intention-driven Lane Change Framework Considering Heterogeneous Dynamic Cooperation in Mixed-traffic Environment

TL;DR

The paper addresses safe and efficient lane changes in mixed-traffic environments by modeling inter-driver heterogeneity through driving-style recognition and cooperation-aware intention inference. It introduces an BC–IRL decision-making framework that fuses intrinsic (style-based) and interactive cooperation cues into a final score $c^{\text{final}}_t$, guiding policy and reward learning, and pairs this with an IRL–MPC motion-planning pipeline to generate collision-free trajectories. Key contributions include a two-stage driving-style module, Learnable Cooperation Score (LCS) and Dynamic Cooperation Score (DCS) with a trainable gate, and an end-to-end training objective combining $\mathcal{L}_{\mathrm{BC}}$, $\mathcal{L}_{\mathrm{IRL}}$, and $\mathcal{L}_{\mathrm{coop}}$, achieving $\text{Accuracy}=0.9418$ and $\text{F1}=0.9427$ on lane-change classification, and showing 4–15% gains over baselines. The results demonstrate improved interpretability, adaptability, and human-like decision-making in complex traffic, with practical impact for context-aware autonomous driving in real-world scenarios.

Abstract

In mixed-traffic environments, where autonomous vehicles (AVs) interact with diverse human-driven vehicles (HVs), unpredictable intentions and heterogeneous behaviors make safe and efficient lane change maneuvers highly challenging. Existing methods often oversimplify these interactions by assuming uniform patterns. We propose an intention-driven lane change framework that integrates driving-style recognition, cooperation-aware decision-making, and coordinated motion planning. A deep learning classifier trained on the NGSIM dataset identifies human driving styles in real time. A cooperation score with intrinsic and interactive components estimates surrounding drivers' intentions and quantifies their willingness to cooperate with the ego vehicle. Decision-making combines behavior cloning with inverse reinforcement learning to determine whether a lane change should be initiated. For trajectory generation, model predictive control is integrated with IRL-based intention inference to produce collision-free and socially compliant maneuvers. Experiments show that the proposed model achieves 94.2\% accuracy and 94.3\% F1-score, outperforming rule-based and learning-based baselines by 4-15\% in lane change recognition. These results highlight the benefit of modeling inter-driver heterogeneity and demonstrate the potential of the framework to advance context-aware and human-like autonomous driving in complex traffic environments.

Figures (8)

• Figure 1: Overall architecture of the proposed intention-driven lane change framework. (i) Driving style recognition provides discrete style labels; (ii) Intention prediction estimates cooperation levels via intrinsic learnable cooperation score (LCS) and interactive dynamic cooperation score (DCS), fused into $c_{\text{final}}$; (iii) Decision-making integrates $c_{\text{final}}$ with vehicle states to predict lane change intentions using policy and reward networks under a BC–IRL training objective; (iv) Motion-planning incorporates IRL-based trajectory prediction of surrounding vehicles and an MPC-based controller to generate safe, executable lane change trajectories.
• Figure 2: Analysis of lane change duration times. (a) The log-normal distribution fit of lane change duration times demonstrates the statistical behavior of lane changes. (b) The residuals plot helps identify any patterns of deviation from the model, indicating the accuracy and fit quality.
• Figure 3: Performance of the lane change start and end point detection algorithm. Spatial distribution of lane change points along vehicle trajectories is illustrated. Red dots denote the initiation points and green dots the termination points. The lane change duration, measured from the start to the end point, reflects the effectiveness and accuracy of the algorithm.
• Figure 4: Analysis of driving-style representation: (a) clustering results show distinct style separation; (b) feature correlation matrix highlights inter-feature dependencies.
• Figure 5: Performance evaluation of the proposed intention-driven lane change decision-making model, including training dynamics, validation metrics, and classification outcomes.