ROAM 理论框架
ROAM 的理论基础：场景分类体系、三层参考架构、8 个评估 KPI。
Layer 1: AI自主响应 (70%)
Layer 2: AI辅助+人工确认 (25%)
Layer 3: 远程驾驶/现场处置 (5%)
A
系统性故障
System-Wide Failure
Anomalies affecting multiple vehicles simultaneously due to infrastructure, cloud, or platform-level failures.
B
感知/决策失效
Perception/Decision Failure
Vehicle's perception system or decision-making logic fails to correctly interpret the driving environment.
C
规划/执行异常
Planning/Execution Anomaly
Vehicle's planning or control system produces dangerous or frozen output despite correct perception.
D
车辆硬件故障
Vehicle Hardware Failure
Anomalies caused by failure of vehicle hardware components — sensors, powertrain, battery, or actuators.
E
外部环境冲突
External Conflict
Anomalies caused by external events or road users that the ADS must respond to.
F
乘客端异常
Passenger-Side Issue
Anomalies originating from or primarily affecting passengers.