Autonomous Driving Decision-making Based on the " Measurement-Semantics-Decision" Space

Objectives: The rapid development of autonomous driving systems has brought forward the challenge of integrating various components from different disciplines. Traditional autonomous driving systems are generally divided into three modules: perception, planning, and control. However, these modules typically handle distinct tasks with limited interaction, resulting in suboptimal performance when encountering complex and diverse traffic scenarios. Additionally, end-to-end autonomous driving systems suffer from a lack of interpretability. This study aims to address these issues by proposing a " Measurement -Semantics-Decision" spatial decision-making chain for autonomous driving. The goal is to complement the pipeline relationship of perception, planning and control of autonomous driving system and to increase the interpretability of end-to-end autonomous driving system. Methods: The autonomous driving "Measurement-Semantics-Decision" space decision chain constructs three spaces: position and geometry, semantic categories, and rules and reasoning. The measurement space provides a unified spatiotemporal reference as well as positional and geometric information of elements in the autonomous driving environment. The semantics space offers a multi-dimensional information representation of element categories, mobility, risk, and rules. The decision space, based on the measurement and semantics spaces, implements functions such as traffic rule transformation, global path planning, local lane planning, guidance line generation, and motion control. Results: This decision chain clarifies the specific tasks within each space of autonomous driving, analyzes the relationships between tasks across spaces, and examines the supporting role of high-definition maps for the three spaces. And then, taking the decision-making of intersection scenarios as an example, the actual application process of this decision-making chain was clarified. Conclusions:The proposed Metric-Semantic-Decision spatial decision-making chain effectively organizes and clarifies the tasks within the perception, planning, and control modules. By structuring the decision-making process into clearly defined spaces, the system is better equipped to handle a broader range of traffic scenarios, including complex and dynamic environments such as intersections. Additionally, the integration of the Metric and Semantic Spaces allows for a more interpretable and explainable decision-making process, addressing the challenges of traditional end-to-end autonomous systems.