Spaces:
Paused
Here is the structured skill file derived from the provided ontology and taxonomy. This format is designed for ingestion by an autonomous AI agent, perception pipeline, or simulation framework.
Skill Definition: Autonomous Driving Semantic Ontology & Taxonomy
Domain Overview
This skill provides a machine-readable vocabulary to describe the driving environment for Automated Driving Systems (ADS) and Advanced Driver Assistance Systems (ADAS).
It transitions validation from mileage accumulation to scenario-based testing.
1. Supported Standards & Frameworks
ISO 34501: Defines the core vocabulary for test scenarios, distinguishing between static scenes and temporal scenarios.
ISO 34502: Establishes a scenario-based safety evaluation framework based on scenario criticality.
ISO 34503: Specifies the Operational Design Domain (ODD), detailing environmental, geometric, and geographic conditions.
ISO 21448 (SOTIF): Identifies rare triggering conditions and functional insufficiencies for the Safety of the Intended Functionality.
ASAM OpenX Suite: Utilizes OpenDRIVE for static road networks, OpenSCENARIO for dynamic behaviors, OpenLABEL for multi-sensor data annotation, and OpenXOntology for semantic relationships.
2. Entity Classification Taxonomy
A. Road Users & Kinematic Profiles
Passenger Vehicles: Passenger cars act as the base model for gap acceptance. SUVs create occlusion zones for forward-facing cameras. Vans have minimal rear visibility. Pickup trucks have variable mass based on payloads. Taxis execute highly unpredictable route behaviors.
Heavy & Commercial Vehicles: Buses block entire lanes during stops. School buses trigger strict longitudinal stopping rules. Heavy trucks require avoidance in lateral blind spots. Semi-trailers mandate extended lateral safety buffers due to trailer sweep.
Emergency Vehicles: Police vehicles, fire trucks, and ambulances possess extreme acceleration or mass and mandate immediate yielding or multi-agent negotiation.
Vulnerable Road Users (VRUs): Pedestrians require continuous pose estimation. Children pose a high risk of occlusion behind parked cars. Wheelchair users require robust LiDAR detection due to atypical radar signatures. Skaters have highly unpredictable trajectory envelopes.
Micro-Mobility: Cyclists require recognition of hand signals for turning. E-bikes feature the visual signature of a bicycle but the acceleration of a scooter.
Animals: Domestic animals execute rapid startle responses. Wildlife herd behavior implies that detecting one animal strongly predicts others. Livestock can completely obstruct roadways.
B. Static Road Objects (The Scenery)
Drivable Surfaces: Includes roads, lanes, shoulders (for minimal risk maneuvers), bike lanes, bus lanes, intersections, roundabouts, highway ramps, and parking spaces.
Physical Constraints & Barriers: Curbs act as hard boundaries for path planners. Sidewalks are the primary domain of VRUs. Guardrails are high-consequence collision objects. Concrete barriers create multipath radar interference.
Vertical Infrastructure: Utility poles serve as reliable localization landmarks. Street lights can cause localized glare or sensor oversaturation.
Pavement Markings: Degradation of lane markings (e.g., from snow) leads to functional insufficiencies. Stop lines trigger mandatory longitudinal deceleration algorithms.
C. Dynamic & Transitory Objects
Uncontrolled Hazards: Falling objects or debris necessitate rapid free-space recalculation.
Temporary Infrastructure: Construction equipment overrides standard HD maps and forces reliance on real-time sensor fusion.
Automated Agents: Delivery robots lack standard human communication cues. Drones confound standard 2D bird's-eye-view perception projections.
3. Activities and Semantic Maneuvers
Vehicle Activities: Rapid braking from a lead vehicle triggers emergency deceleration protocols. Open doors instantly redefine drivable free space.
Pedestrian Activities: Walking speed determines the predicted intersection point with the ego vehicle's path. Jaywalking triggers immediate emergency deceleration.
Traffic Control Elements: Regulatory signs (e.g., Stop, Yield) enforce mandatory constraints. Variable message signs (VMS) require complex natural language processing (NLP) to interpret. Flaggers and police gestures supercede all static lights and signs.
4. Operational Design Domain (ODD) Context
Weather & Visibility: Rain and snow distort camera lenses and obscure lane markings. Fog and smoke scatter LiDAR pulses, forcing heavier reliance on radar.
Surface & Force: Ice reduces tire-road friction, expanding required stopping distances. High crosswinds require compensatory steering torque.
Lighting: Dawn and dusk create severe horizontal solar glare, which is a classic SOTIF triggering condition.
5. Semantic Ontology Graph
The system utilizes ontological mapping to reason about physical and legal dependencies.
Vehicle -> is_cutting_into -> Lane: Triggers deceleration to restore safety margins.
Pedestrian -> intends_to_cross -> Crosswalk: Mandates yielding the right-of-way.
Weather (Ice) -> degrades -> Road (Surface): Updates global friction coefficient variables.
Police Officer -> overrides -> Traffic Light: Forces the system to ignore the traffic light state.
6. Machine-Readable Integration Schema (ASAM OpenLABEL)
The following JSON schema can be directly ingested into your project's perception pipeline to model SOTIF corner cases, such as a pedestrian crossing in heavy rain.
{
"openlabel": {
"metadata": {
"schema_version": "1.0.0",
"scenario_name": "Urban_Pedestrian_Crossing_Rain_SOTIF",
"description": "Ego vehicle encounters a jaywalking pedestrian in heavy rain, triggering an RSS braking maneuver.",
"ontologies": {
"asam_openx": "https://www.asam.net/standards/asam-openxontology"
}
},
"contexts": {
"env_001": {
"name": "Heavy_Rain",
"type": "EnvironmentalCondition",
"attributes": {
"precipitation_rate": "15.0 mm/hr",
"visibility_reduction": true,
"surface_friction_coefficient": 0.5
}
},
"odd_001": {
"name": "Urban_Arterial",
"type": "OperationalDesignDomain",
"attributes": {
"speed_limit": "50 km/h",
"road_type": "CityStreet",
"region": "US_RightHandDrive"
}
}
},
"objects": {
"ego_vehicle": {
"name": "Ego",
"type": "Vehicle",
"sub_type": "PassengerCar",
"kinematics": {
"velocity_x": 12.5,
"acceleration_x": -4.0
}
},
"ped_001": {
"name": "Pedestrian_Adult",
"type": "VulnerableRoadUser",
"sub_type": "Pedestrian",
"attributes": {
"pose": "Walking",
"is_occluded": false
}
},
"infra_001": {
"name": "Main_Crosswalk",
"type": "StaticRoadObject",
"sub_type": "Crosswalk"
}
},
"actions": {
"act_001": {
"type": "Emergency_Braking",
"actor": "ego_vehicle",
"trigger": "rss_longitudinal_distance_violation"
},
"act_002": {
"type": "Jaywalking",
"actor": "ped_001",
"description": "Pedestrian crossing outside designated crosswalk infra_001."
}
},
"relations": {
"rel_001": {
"type": "interacts_with",
"subject": "ego_vehicle",
"object": "ped_001"
},
"rel_002": {
"type": "modifies",
"subject": "env_001",
"object": "ego_vehicle",
"description": "Rain reduces optimal braking friction, increasing required RSS stopping distance."
}
},
"frames": {
"0": {
"timestamp": 1690384512.001,
"frame_properties": {
"sotif_triggering_condition": "High_Rain_Glare"
},
"object_data": {
"ped_001": {
"bbox_3d": [15.2, 3.1, 0.0, 0.6, 0.6, 1.7, 1.57]
}
}
}
}
}
}