HATIS: A Hierarchical Autonomous Traffic Intelligence System — Capstone Project Report

Engineering notes

End-to-end simulation, CARLA validation, benchmarking, and real-world testing are identified as future work.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

This capstone project report presents HATIS (Hierarchical Autonomous Traffic Intelligence System), a proposed modular architecture for explainable and verifiable autonomous driving. The report comprises two implemented analytical components and a proposed autonomous-driving architecture. Part I develops a CNN-based vehicle object-detection pipeline for road-user detection and localisation. Part II performs exploratory analysis of publicly reported Tesla Autopilot-related accident data. Building on these components, the report proposes a seven-layer HATIS architecture comprising Digital Road Intelligence, Neural Perception, 3D Dynamic World Modeling, Traffic Population Intelligence, a Mathematical Decision Engine, Vehicle Dynamics and Control, and Continuous Learning and Validation. A central architectural contribution is the proposed Mathematical Decision Engine and its Triple-Verification Autonomous Driving Framework (TVADF), in which candidate driving decisions are subjected to AI, geometric, and mathematical safety verification before execution. The architecture also introduces nested immediate, short-term, and route-level predictive safety horizons and an explicit human-readable decision trail. The report presents HATIS as a research and architectural proposal rather than a completed autonomous-driving implementation. End-to-end simulation, CARLA validation, benchmarking, and real-world testing are identified as future work. Scope and status The vehicle-detection and accident-data-analysis components constitute the principal analytical work of the capstone. The broader HATIS architecture is presented as a proposed research framework. The proposed Layers 0 and 3–6 were not implemented end-to-end or tested in a simulation environment within the scope of this capstone.

Links and sources

Comments