generated: '2026-08-30'
method: derived
source: >-
  json-schema/ai-habitat-simulator-config-schema.json, json-schema/ai-habitat-agent-config-schema.json,
  json-schema/ai-habitat-sensor-spec-schema.json, json-schema/ai-habitat-observation-schema.json,
  json-schema/ai-habitat-agent-observation-schema.json, json-schema/ai-habitat-episode-schema.json,
  json-schema/ai-habitat-navigation-goal-schema.json, json-schema/ai-habitat-task-config-schema.json,
  vocabulary/ai-habitat-vocabulary.yaml
provider: AI Habitat
providerId: ai-habitat
description: >-
  Entity-relationship graph for the AI Habitat configuration and observation model, derived from the
  JSON Schemas already held in this repository. There is no OpenAPI to derive from — AI Habitat is an
  in-process simulator, so these "entities" are Python/C++ configuration objects and per-step
  observation payloads, not REST resources. Relationships below are read from the id-reference and
  embedded-array fields present in those schemas; nothing is inferred beyond what the schemas state.
caveat: >-
  The source schemas are API Evangelist-derived artifacts, not provider-published contracts. This data
  model inherits that provenance: it is a faithful reading of our own schemas, which were in turn
  written from the Habitat documentation at https://aihabitat.org/docs/.
entities:
- name: SimulatorConfig
  schema: json-schema/ai-habitat-simulator-config-schema.json
  description: Top-level simulator instantiation config — scene, physics, device and determinism.
  key_fields: [scene_id, default_agent_id, physics_simulation_library, gpu_device_id, random_seed, enable_physics]
- name: AgentConfig
  schema: json-schema/ai-habitat-agent-config-schema.json
  description: Embodiment of one agent — physical extents, sensor rig and action space.
  key_fields: [height, radius, body_type, action_space, sensor_specifications]
- name: SensorSpec
  schema: json-schema/ai-habitat-sensor-spec-schema.json
  description: One sensor mounted on an agent, addressed by uuid.
  identifier: uuid
  key_fields: [uuid, sensor_type, resolution, position, orientation, hfov]
- name: Observation
  schema: json-schema/ai-habitat-observation-schema.json
  description: Raw per-step sensor readings returned by the simulator.
  key_fields: [rgb, depth, semantic, gps, compass]
- name: AgentObservation
  schema: json-schema/ai-habitat-agent-observation-schema.json
  description: Task-shaped observation dictionary handed to a policy, keyed by sensor uuid.
  key_fields: [rgb, depth, pointgoal_with_gps_compass, objectgoal]
- name: Episode
  schema: json-schema/ai-habitat-episode-schema.json
  description: One instantiated task instance — scene, agent start pose and goals.
  identifier: episode_id
  key_fields: [episode_id, scene_id, start_position, start_rotation, goals, info]
- name: NavigationGoal
  schema: json-schema/ai-habitat-navigation-goal-schema.json
  description: A goal location with a success radius.
  key_fields: [position, radius]
- name: TaskConfig
  schema: json-schema/ai-habitat-task-config-schema.json
  description: Task definition — type, episode budget, measurements and action space.
  key_fields: [type, max_episode_steps, measurements, actions, goal_sensor_uuid]
relationships:
- from: SimulatorConfig
  to: SensorSpec
  type: has_many
  via: sensor_specifications
- from: AgentConfig
  to: SensorSpec
  type: has_many
  via: sensor_specifications
- from: SensorSpec
  to: Observation
  type: has_one
  via: uuid
  note: Each sensor's uuid becomes the key its reading appears under in the observation dictionary.
- from: Episode
  to: NavigationGoal
  type: has_many
  via: goals
- from: Episode
  to: SimulatorConfig
  type: belongs_to
  via: scene_id
  note: >-
    Soft reference — Episode.scene_id and SimulatorConfig.scene_id name the same scene asset path;
    there is no scene entity with its own identifier in the schema set.
- from: TaskConfig
  to: SensorSpec
  type: belongs_to
  via: goal_sensor_uuid
  note: goal_sensor_uuid resolves against SensorSpec.uuid.
- from: TaskConfig
  to: AgentObservation
  type: has_one
  via: actions
  note: >-
    The task's action space and measurement list determine which keys the policy sees in
    AgentObservation; the schemas record the fields, not the binding logic.
identifiers:
- entity: Episode
  field: episode_id
  form: opaque string
  example: ep_001
- entity: SensorSpec
  field: uuid
  form: lowercase sensor name
  example: rgb
- entity: TaskConfig
  field: type
  form: '<TaskName>-v<N>'
  example: ObjectNav-v1
- entity: SimulatorConfig
  field: scene_id
  form: filesystem path to a .glb/.ply scene asset
  example: data/scene_datasets/habitat-test-scenes/apartment_1.glb
counts:
  entities: 8
  relationships: 7