DA-MORL: Structural-Analysis-Coupled Reinforcement Learning for Building Demolition Sequencing (code and data)
Year: 2026
Type: software
Abstract: Code, trained models, and evaluation data accompanying the manuscript Building demolition sequence optimization using structural-analysis-coupled deep reinforcement learning and a CAD-to-graph representation (submitted to the Journal of Computational Design and Engineering; an earlier version was submitted under the title "An AI- and CAD-Based Building Demolition Sequence Optimization Model Combining Structural Analysis and Reinforcement Learning"). Includes the OpenSees-coupled Gymnasium environment, MaskablePPO training scripts, baselines, and all experiment scripts and per-seed results reported in the paper. Version 3 (2026-09-19): DAMORL_code_release_v2.zip supersedes the v1 archive (same damorl-code/ layout) and adds the CAD-to-MRG worked examples (extracted and reference member relation graphs, reading protocol, conversion and scoring scripts), the revision experiments (floor-access constraint, lateral-load replays, load-persistence sensitivity, budget-matched optimizers, clean-environment reproduction of Table 3), four additional trained policies, and a README describing the changes. The supplement archive deposited in version 2 is included in this archive. Version 2 (2026-09-19): supplement archive only. Version 1 (2026-08-26): code, 24 trained policies, per-seed evaluation data. Source drawings are not redistributable; see README.md inside the archive. License: MIT (code), CC BY 4.0 (data tables).
Authors: Kim, Young-Hwan, Ko, Young-Hee
Institution: Seoul School of Integrated Sciences and Technologies
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204226986