PGM-AUA V1.0 Data Types – Visual Scene Enhancers

(Tentative)

Definition

Functional Requirements

• Semantic Enrichment
  • • Must transform baseline Visual Scene Descriptors (VSD0) into enriched structures (VSD1→VSD2→VSD3) that add relations, salience, occlusion, affordances, and reachability needed by downstream AIMs.
    • Must preserve and normalise original object attributes (IDs, types, bounding data) while appending domain-aware annotations.
  • Directive Alignment
    • Must encode the active Visual Action Directive (VAD)—including priorities, constraints, and scope—so enrichment remains goal-aligned across the loop with DAC and VSR.
    • Must support conditional logic that binds enrichment to directive classes and parameters, mirroring the intent/domain-class consistency pattern in the Intent data-type page.
  • Spatial Reasoning Features
    • • Must provide refined localisation (pose, depth/occlusion state, proximity zones), affordance inference, and salience mapping usable by Prompt Creation and A-User Formation AIM.
      • Should expose confidence scores per inference (e.g., occlusion, affordance) to enable downstream filtering and fallback, following the confidence pattern used in Intent.
  • Object & Relation Graph
    • Must represent inter-object relations (e.g., adjacency, containment, support, collision risk) with time-stamped updates for dynamic scenes.
    • Should include semantic tags (domain-specific labels) for objects and zones to improve reasoning fidelity, consistent with other PGM data-type semantics.
  • Constraints & Feasibility
    • Must report feasibility flags (e.g., reachable, valid, occluded levels) and residual constraints not satisfied during enrichment, enabling conformance decisions.
    • Must capture directive- and domain-derived constraints injected by DAC (e.g., prohibited zones, safety margins, governance limits).
  • Resolution Hints & Rendering Guidance
    • Must provide Resolution Hints (e.g., preferred viewpoints, LOD prioritisation, update cadence) to assist rendering and prompt planning.
    • Should include alternative viewpoints and fallback hint sets when confidence or feasibility drops below thresholds.
  • Traceability & Governance
    • Must carry a Trace block (originating module, timestamps aligned to M-Instance time base, version), supporting audit and provenance across enrichment cycles.
    • Should encode governance metadata (policy references, compliance status, decision path) to satisfy PGM conformance and potential cross-standard governance needs.
  • Identity & Referencing
    • Must include globally unique IDs for the VSE record, referenced objects/zones, and linkage to M-Instance/M-Environment identifiers to enable chaining and logging.
    • Should support cross-referencing to other data types (e.g., Intent, Context, Prompt Plans) for end-to-end workflow integration.
  • Performance Hooks
    • Should expose metrics fields for enrichment latency, coverage (percentage of scene enriched), and uncertainty summaries to feed performance sections of the spec.

  Syntax

  https://schemas.mpai.community/PGM1/V1.0/data/VisualSceneEnhancers.json

  Semantics