Agentic Simulation Foundry
Does a generate → retrieve → mutate → execute → score pipeline transfer from autonomous driving to logistics, GIS, or facility-operations scenarios?
Hypothesis
PlannerForge's scenario-generation loop is domain-agnostic enough to swap driving scenarios for a logistics or GIS domain without redesigning the core pipeline.
Background
PlannerForge (arXiv 2609.08965, accepted EMNLP 2026) reports 193/200 executable scenarios vs. 144 for a baseline, and planner success rising from 50.4% to 70.2% with grounded scenario generation — with a verified official repository.
Method
Select a non-driving domain (logistics or facility operations); adapt PlannerForge's tool adapters; measure scenario validity and diversity against the original benchmark.
Measurement Plan
Scenario validity, diversity, retrieval quality, and edit-success rate in the new domain.
Expected Artifact
A domain-neutral simulation pipeline, demonstrated on at least one non-driving domain.
Limitations
The original paper is scoped to autonomous driving; domain transfer is an open research question, not a guaranteed result.
Status
This experiment is Proposed — queued, not started. Results, data, and measurements will appear here once it actually runs.
