AGV Safety-Acceptance Simulation Benchmark: Code and Data for "Mining Safety-Acceptance Rules from Kinematic AGV Simulations"

Code, simulated data and results accompanying the article “Mining Safety-Acceptance Rules from Kinematic AGV Simulations: Surrogate-Assisted Active Learning for Parametric Design of Automated Guided Vehicles” (submitted to Algorithms, MDPI). The package provides a planted-structure benchmark for data mining of simulation data: a 13-parameter parametric model of an automated guided vehicle (AGV), a simplified 2D kinematic warehouse simulation model with reactive pedestrians, test pieces and a front safety laser scanner, and three warehouse scenarios. The corpora contain 900,000 simulated episodes in 12 independent folds; each design is labeled from physical events (incidents, availability failures) over five stochastic episodes. Contents: code/: simulator and scripts for experiments E1–E10 (surrogate classifiers, active learning, rule mining with CART/PRIM/association rules, robustness, design repair incl. CMA-ES and constrained Bayesian optimization, interpretable models incl. GA2M). python code/run_all.py reproduces all results with fixed seeds. data/: simulated corpora (Parquet) with data dictionary. results/: raw and summary results (CSV/JSON), method details (METHODS.md) and the pre-specified analysis plan for E6–E10 (V3_PLAN.md). figures/: all figures (300 dpi). The simulator is a simplified kinematic model, not a validated digital twin, and the results do not constitute evidence of conformity with ISO 3691-4 or any other standard.