This research line builds ontological infrastructure for automated planning and explainability. It starts with a general planning ontology (PO) for planner behavior, expands into multi-agent execution semantics (maPO), and culminates in OMEGA, an explainability pipeline that converts raw MAPF traces into human-readable causal narratives.
The unifying goal is to make planning artifacts queryable and reusable: selecting planners from historical signals, extracting reusable macro patterns, and explaining behavior in robot-deployment settings where standard solver logs are hard to interpret.
CODS 2024
Planning Ontology Foundation
Initial ontology schema for planning domains, planners, tasks, and performance properties.
Discover Data 2025
Expanded PO + Applications
Extended ontology coverage and evidence on planner selection and planning knowledge reuse.
AAAI-MAKE 2026
maPO for MAPF
Ontology extension for collision events, replanning causes, and multi-agent execution semantics.
AAAI 2026 Demo
OMEGA Explainability Tool
Operational system that maps MAPF execution logs to contextual, user-facing explanations.
Core Thesis: planning ontologies are not just metadata repositories; they are decision-support layers that improve planner choice, execution understanding, and human trust.
Key Idea: PO is an OWL 2 ontology that organizes planning domains, solver properties, run-time outcomes, and PROV-O provenance into a SPARQL-queryable knowledge graph.
Planner Performance & Selection
Historical IPC traces become structured evidence for selecting planners that best match domain and instance characteristics before execution.
Macro Actions Extraction
Provenance-rich action traces support mining reusable macro operators that reduce search burden in related planning tasks.
Tooling Integration
A Planning.Domains plugin converts PDDL artifacts to RDF, enabling direct ontology-backed analysis in existing planning workflows.
PO schema linking planners, domains, tasks, and execution outcomes.
Planning.Domains plugin for PDDL to RDF translation.
maPO & OMEGA: maPO models MAPF-specific execution semantics including agent conflicts, collision events, and replanning context. OMEGA uses this schema to convert low-level trace logs into causal, user-readable explanations.
Key entities include ma:Agent, ma:CollisionEvent, ma:ConflictAlert, and ma:ReplanningStrategy, connected through provenance links that preserve event causality.
OMEGA Framework: execution traces are transformed to RDF graphs over maPO, queried through SPARQL templates, and rendered as contextual explanations aligned with animation states. In user studies, participants preferred OMEGA explanations over raw logs (95.2% preference; 4.40/5 clarity).
Log-to-graph pipeline and interactive OMEGA explanations.
The demo emphasizes practical explainability during MAPF execution, showing why collisions occur, how conflict alerts are triggered, and which replanning strategy resolves each issue.
Representative Publications
AAAI '26 DemonstrationOMEGA: An Ontology-Driven Tool for Explaining Multi-Agent Path Finding
