Complex but solvable: towards a cognitive human-like risk-identification model for AV-HV mixed traffic.

Area of Science:

• Artificial Intelligence
• Autonomous Driving Systems
• Traffic Safety

Background:

• Human-driven vehicles (HVs) present behavioral uncertainty in mixed traffic, challenging autonomous vehicles (AVs) with risk identification, especially in dynamic interweaving zones.
• Current risk identification methods lack generalization, responsiveness to extreme risks, and interpretability in complex mixed traffic scenarios.

Purpose of the Study:

• To develop a cognitive human-like risk-identification model (CHRIM) inspired by human adaptive learning and consultation behaviors.
• To improve AV risk identification accuracy and robustness in interweaving areas with high traffic dynamics and complex interactions.

Main Methods:

• Constructed a cognitive optimization method based on the human behavior-based optimization (HBBO) algorithm.
• Developed the Cognitive Human-like Risk-Identification Model (CHRIM) simulating human driver cognitive abilities like abstract understanding, retrospective reasoning, and strategy optimization.
• Validated the model using real trajectory data from an urban expressway interweaving area.

Main Results:

• The CHRIM achieved superior performance in risk identification accuracy (ACC: 0.9543), robustness (MCC: 0.8906), and overall effectiveness (KAP: 0.8782).
• Outperformed mainstream methods including fine tree model (FTM), random subspace model (RSM), efficient logistic regression model (ELRM), and neural networks (NN).
• An explainable analysis framework provided intuitive insights into the AV's human-like risk identification and risk evolution process.

Conclusions:

• The proposed CHRIM effectively addresses limitations of existing methods by incorporating human-like cognitive processes for adaptive risk identification.
• The model demonstrates significant improvements in accuracy and robustness for AVs interacting with human-driven vehicles in complex interweaving scenarios.
• The explainable framework enhances understanding of AV decision-making in critical traffic situations.