PRISM：面向自动驾驶的智能体多模型主动安全架构
原标题：PRISM: An Agentic Multi-Model Architecture for Proactive Safety in Autonomous Transportation Systems
AI 摘要
PRISM 研究提出了一种名为 PRISM 的智能体多模型安全架构，用于自动驾驶系统，将安全从被动碰撞避免转变为主动持续风险管理。该系统结合逆碰撞概率建模、三个专用模型（轨迹、天气、VRU交互）以及强化学习协调层，提供四级安全干预。在自然驾驶数据集上验证，平均安全评分为68，77.6%场景为咨询级，3.8%为近碰撞率，约11%场景升级为干预或紧急响应。该框架旨在减少城市环境中弱势道路使用者（VRU）的风险，并为车辆制造商、车队运营商和城市规划者提供可解释的安全指标。
正文节选
captionUnknown document class \NameTagRoy, July 29, 2026 PRISM: AN AGENTIC MULTI-MODEL ARCHITECTURE FOR PROACTIVE SAFETY IN AUTONOMOUS TRANSPORTATION SYSTEMS Abstract Autonomous and intelligent transportation systems operate within complex urban environments, where safety relies on interactions among vehicle behavior, environmental conditions, and vulnerable road users (VRUs), including pedestrians and cyclists. Most advanced driver assistance systems (ADAS) primarily employ reactive mechanisms,