Robust Carbon-Budget Screening for Offline Joint Inventory and Freight Control Under Emission-Factor Shift

Authors

  • Adrian Stewart School of Computer Science, College of Computing, Georgia Institute of Technology, USA
  • Eva Collins School of Computer Science, College of Computing, Georgia Institute of Technology, USA

DOI:

https://doi.org/10.54097/fd8ara21

Keywords:

Offline reinforcement learning, carbon budget, inventory and transportation, robust optimization, action projection, emission factor uncertainty

Abstract

Offline inventory and freight policies are increasingly evaluated under finite carbon budgets, yet a predicted-emissions filter does not itself prevent over-budget realized emissions when vehicle factors are uncertain. Building on the action-level pathwise carbon screen of carbon-aware conservative Q-learning (CA-CQL), which guarantees compliance for a predicted emissions ledger, we extend the screen to an uncertainty-bound feasibility projection that preserves behavior-supported replenishment proposals. The study constructs synthetic fixed logs for a supplier, distribution center, and three retailers and evaluates policy-independent demand and emission-error streams. Five offline training datasets contain 24,000 transitions in total, and each policy is evaluated over 250 held-out trajectories in each uncertainty condition. Under a 4,300 kg CO2e cap and emission multipliers in [0.8, 1.2], unfiltered behavior violates the cap in 30.0% of episodes, the nominal CA-CQL screen in 5.2%, and a 1.20 robust screen in 0%, while demand fill declines from 89.64% to 89.50%. Under mild [0.9, 1.3] and severe [1.1, 1.5] shifts, the 1.20 screen has 2.0% and 18.8% observed violations; thus the conditional guarantee does not extend past its assumed bound. A 1.35 factor reduces severe-shift violations to 6.4%. Comparisons with behavior cloning and support-regularized offline Q-learning separate constraint feasibility from policy quality, while common-random-number evaluation and seed-level paired tests quantify uncertainty. The result is a reproducible, auditable robust-dispatch interface rather than a claim of end-to-end optimal inventory reinforcement learning.

Downloads

Download data is not yet available.

References

[1] Clark, A. J., & Scarf, H. (1960). Optimal policies for a multi-echelon inventory problem. Management Science, 6(4), 475–490. https://doi.org/10.1287/mnsc.6.4.475 DOI: https://doi.org/10.1287/mnsc.6.4.475

[2] Campbell, A. M., Clarke, L., Kleywegt, A. J., & Savelsbergh, M. W. P. (1998). The inventory routing problem. In Fleet Management and Logistics (pp. 95–112). Springer. https://doi.org/10.1007/978-1-4615-5755-5_4 DOI: https://doi.org/10.1007/978-1-4615-5755-5_4

[3] Li, P., Bian, M., & Lin, Y. (2026). Carbon-aware offline reinforcement learning for joint inventory replenishment and transportation scheduling in multi-echelon supply chains. Highlights in Business, Economics and Management, 69, 251–262. https://doi.org/10.54097/p40ecd94 DOI: https://doi.org/10.54097/p40ecd94

[4] Kumar, A., Zhou, A., Tucker, G., & Levine, S. (2020). Conservative Q-learning for offline reinforcement learning. Advances in Neural Information Processing Systems, 33, 1179–1191.

[5] Fujimoto, S., Meger, D., & Precup, D. (2019). Off-policy deep reinforcement learning without exploration. Proceedings of the International Conference on Machine Learning, 2052–2062.

[6] Kumar, A., Fu, J., Soh, M., Tucker, G., & Levine, S. (2019). Stabilizing off-policy Q-learning via bootstrapping error reduction. Advances in Neural Information Processing Systems, 32.

[7] Fu, J., Kumar, A., Nachum, O., Tucker, G., & Levine, S. (2020). D4RL: Datasets for deep data-driven reinforcement learning. arXiv preprint arXiv:2004.07219.

[8] Altman, E. (1999). Constrained Markov Decision Processes. CRC Press.

[9] Achiam, J., Held, D., Tamar, A., & Abbeel, P. (2017). Constrained policy optimization. Proceedings of the International Conference on Machine Learning, 22–31.

[10] Bektaş, T., & Laporte, G. (2011). The pollution-routing problem. Transportation Research Part B, 45(8), 1232–1250. https://doi.org/10.1016/j.trb.2011.02.004 DOI: https://doi.org/10.1016/j.trb.2011.02.004

[11] Demir, E., Bektaş, T., & Laporte, G. (2014). A review of recent research on green road freight transportation. European Journal of Operational Research, 237(3), 775–793. https://doi.org/10.1016/j.ejor.2013.12.033 DOI: https://doi.org/10.1016/j.ejor.2013.12.033

[12] Cheng, C., Qi, M., Wang, X., & Zhang, Y. (2016). Multi-period inventory routing problem under carbon emission regulations. International Journal of Production Economics, 182, 263–275. https://doi.org/10.1016/j.ijpe.2016.09.001 DOI: https://doi.org/10.1016/j.ijpe.2016.09.001

[13] Kleywegt, A. J., Nori, V. S., & Savelsbergh, M. W. P. (2002). The stochastic inventory routing problem with direct deliveries. Transportation Science, 36(1), 94–118. https://doi.org/10.1287/trsc.36.1.94.574 DOI: https://doi.org/10.1287/trsc.36.1.94.574

[14] Kleywegt, A. J., Nori, V. S., & Savelsbergh, M. W. P. (2004). Dynamic programming approximations for a stochastic inventory routing problem. Transportation Science, 38(1), 42–70. https://doi.org/10.1287/trsc.1030.0041 DOI: https://doi.org/10.1287/trsc.1030.0041

[15] Campbell, A. M., & Savelsbergh, M. W. P. (2004). A decomposition approach for the inventory-routing problem. Transportation Science, 38(4), 488–502. https://doi.org/10.1287/trsc.1030.0054 DOI: https://doi.org/10.1287/trsc.1030.0054

[16] Trudeau, P., & Dror, M. (1992). Stochastic inventory routing: Route design with stockouts and route failures. Transportation Science, 26(3), 171–184. https://doi.org/10.1287/trsc.26.3.171 DOI: https://doi.org/10.1287/trsc.26.3.171

[17] Sutton, R. S., & Barto, A. G. (2018). Reinforcement Learning: An Introduction (2nd ed.). MIT Press.

[18] World Resources Institute & World Business Council for Sustainable Development. (2011). Corporate Value Chain (Scope 3) Accounting and Reporting Standard.

[19] Smart Freight Centre. (2024). Global Logistics Emissions Council Framework for Logistics Emissions Accounting and Reporting (Version 3.1).

[20] UK Department for Energy Security and Net Zero & Department for Environment, Food and Rural Affairs. (2026). Greenhouse Gas Reporting: Conversion Factors 2026.

[21] Gibbs, I., & Candès, E. (2021). Adaptive conformal inference under distribution shift. Advances in Neural Information Processing Systems, 34.

[22] Dong, Q., & Yuan, Y. (2025). Data-driven distributionally robust supplier selection and order allocation problems considering carbon emissions. International Transactions in Operational Research, 32(2), 1119–1145. https://doi.org/10.1111/itor.13328 DOI: https://doi.org/10.1111/itor.13328

Downloads

Published

19-11-2026

Issue

Section

Articles

How to Cite

Stewart, A., & Collins, E. (2026). Robust Carbon-Budget Screening for Offline Joint Inventory and Freight Control Under Emission-Factor Shift. International Journal of World Economic Research, 4(1), 51-57. https://doi.org/10.54097/fd8ara21