QDRL TurtleBot Reproducibility Archive

This record contains the data and reproducibility materials for a computationally constrained partial replication and training-design analysis of quantum deep reinforcement learning for simulated TurtleBot navigation. The experiments compare a classical neural-network DDQN function approximator with two parameterized quantum-circuit configurations, PQC-1 and PQC-3, across three static navigation environments (3 × 3, 4 × 4, and 5 × 5) and one dynamic 12 × 12 LiDAR environment. The additional analysis examines how the initial replay-buffer collection size, INITIAL_COLLECT, is associated with PQC-3 learning outcomes. The record contains two principal archives: qdrl_paper_data_378_runs_2026-07-24.tar.gz: Raw training-result data comprising 378 physical pickle files and 358 unique run contents. These include the replication cohort, the systematic INITIAL_COLLECT experiments, neural-network reference runs, and exploratory INITIAL_COLLECT = 20,000 experiments. qdrl_turtlebot_reproducibility_v1.0.0_2026-07-30.tar.gz: Processed datasets, the numbered analysis pipeline, publication-ready tables and figures, validation logs, experiment-specific configuration files, dependency records, Git provenance, source-code snapshots, audit documentation, licence information, and SHA-256 manifests. The reproducibility archive includes source snapshots derived from the qdrl-turtlebot-env and qdrl-turtlebot-eval repositories associated with Hohenfeld et al., “Quantum Deep Reinforcement Learning for Robot Navigation Tasks,” IEEE Access, 2024, DOI: 10.1109/ACCESS.2024.3417808. Their original BSD 3-Clause licences and copyright notices are retained. Machine-specific paths and identifiers were removed from the public-release copy. The original protected archive was retained separately, and all public files were validated using SHA-256 manifests. The analysis archive also passed a complete extraction and round-trip hash validation. Original analysis code and authored source modifications are released under the BSD 3-Clause License. Original data products, derived tables, figures, logs, and documentation are released under the Creative Commons Attribution 4.0 International licence, except where separate third-party ownership or licensing is stated.