{
  "field_id": "F022",
  "field": "autonomous driving",
  "directions": [
    {
      "direction_id": "G0",
      "name": "Perception of the surrounding driving scene",
      "scope": "Identify the surrounding driving scene from onboard observations or observations shared across vehicles and roadside infrastructure, including cooperative perception, 3D object detection, present spatial semantics and occupancy representations, and drivable-area and road-scene segmentation."
    },
    {
      "direction_id": "G1",
      "name": "Recovery of road geometry and structure",
      "scope": "Recover lane and road structure from driving observations, including local lane shapes, map elements, vectorized road maps, and connectivity among lanes and traffic elements."
    },
    {
      "direction_id": "G2",
      "name": "Vehicle localization against observed and mapped geometry",
      "scope": "Determine vehicle position and pose through place recognition, sensor-to-map matching, or geometric mapping developed jointly with localization."
    },
    {
      "direction_id": "G3",
      "name": "Tracking and forecasting traffic dynamics",
      "scope": "Use observations over time to maintain traffic participants’ identities and motion states and to predict their future behavior or changes in occupied scene space."
    },
    {
      "direction_id": "G4",
      "name": "Learning and selecting driving actions",
      "scope": "Learn or select driving behavior in road space shared with other traffic participants, including single-vehicle policies, ego-vehicle planning that accounts for predictions and uncertainty, and coordination of multiple vehicles’ passage."
    },
    {
      "direction_id": "G5",
      "name": "Generation and predictive modeling of driving worlds",
      "scope": "Construct synthetic environments, traffic, and sensor observations for driving research, or learn future driving-scene states and possible action-conditioned outcomes."
    },
    {
      "direction_id": "G6",
      "name": "Driving-system testing and safety validation",
      "scope": "Use challenging scenarios, requirements, and error models in virtual testing to expose driving- or perception-system failures and assess performance and safety."
    },
    {
      "direction_id": "G7",
      "name": "Perception under distribution shift and anomalous conditions",
      "scope": "Study perception reliability beyond ordinary training conditions, including transfer across specified environmental or data-domain changes and recognition of anomalous inputs, detection outliers, and rare road hazards."
    },
    {
      "direction_id": "G8",
      "name": "Language-grounded driving reasoning and interaction",
      "scope": "Connect language with driving scenes, actions, and intentions, both to produce driving decisions from language instructions and to evaluate scene understanding and reasoning expressed through question answering or language explanations."
    }
  ]
}