Reference Architecture

Three-layer decision model for handling L4+ robotaxi operational anomalies. Core principle: AI handles routine anomalies autonomously; humans handle the exceptions.

Layer 1: AI Autonomous Response

AI自主响应

70%

Latency: < 2 secHuman role: None

Layer 2: AI-Assisted + Human Confirm

AI辅助+人工确认

25%

Latency: < 60 secHuman role: Confirm / Override

Layer 3: Remote Driving / On-Site

远程驾驶/现场处置

Latency: < 5 minHuman role: Direct control

Scenario-to-Layer Mapping

ID	Scenario	Primary	Escalation
A1	Cloud/Network Mass Failure	L1	L3
A2	OTA Update Failure	L2	L3
A3	External Infrastructure Failure	L2	L3
A4	Server Decision System Crash	L1	L3
B1	Complex Intersection Hesitation	L1	L2
B2	Object Misidentification	L1	L2
B3	Extreme Weather Perception Degradation	L1	L2
B4	Abnormal Human Behavior	L1	L2
B5	GPS/Localization Drift	L1	L2
C1	Mid-Road Freeze	L2	L3
C2	Unexpected Hard Braking	L1	—
C3	Dangerous Lane Change	L1	—
C4	Static Object Collision	L1	L3
C5	Low-Speed Pedestrian/Cyclist Collision	L1	L3
D1	Sensor Hardware Failure	L1	L2
D2	Powertrain/Battery Failure	L2	L3
D3	Vehicle Fire	L3	—
D4	Brake/Steering Actuator Failure	L1	L3
E1	Struck by Other Vehicle	L1	L3
E2	Road Construction/Closure	L2	L3
E3	Emergency Vehicle Interaction	L2	L3
E4	Police/Traffic Officer Gesture Failure	L2	L3
F1	Passenger Trapped in Vehicle	L2	L3
F2	Dangerous Pickup/Dropoff Location	L1	L2
F3	Passenger Interference with Vehicle	L2	L3
F4	Passenger Medical Emergency	L3	—

Urgency-Severity Matrix

	U0 Low	U1 Medium	U2 High	U3 Immediate
S0 Near-Miss	Routine	Monitor	Unusual	Investigate
S1 Minor	Log	Resolve	Priority	Priority
S2 Moderate	Review	Priority	Critical	Critical
S3 Severe	—	Critical	Critical	Emergency
S4 Critical	—	—	Emergency	Emergency