Publications

Generalizing Beyond Suboptimality: Offline Reinforcement Learning Learns Effective Scheduling through Random Solutions

Published in Transactions on Machine Learning Research, 2026

CDQAC is an offline RL algorithm for job shop and flexible job shop scheduling that learns effective policies from historical data without any simulator. It matches or beats state-of-the-art online and offline RL baselines from only 10–25 training instances, generalizes to instances ten times larger than those seen in training, and performs best when trained on solutions generated by a random heuristic.

Project page

Recommended citation: van Remmerden, J., Bukhsh, Z., & Zhang, Y. (2026). Generalizing Beyond Suboptimality: Offline Reinforcement Learning Learns Effective Scheduling through Random Solutions. Transactions on Machine Learning Research.
Download Paper

Offline Reinforcement Learning for Learning to Dispatch for Job Shop Scheduling

Published in Machine Learning, 2025

Offline Learned Dispatching (Offline-LD) is the first fully end-to-end offline RL approach for the Job Shop Scheduling Problem, learning dispatching policies from historical schedules with no simulator at all. Trained on only 100 constraint-programming solutions, it outperforms online RL baselines that require millions of simulated interactions — and performs no worse when that training data is deliberately corrupted with noise.

Recommended citation: van Remmerden, J., Bukhsh, Z., & Zhang, Y. (2025). Offline reinforcement learning for learning to dispatch for job shop scheduling. Machine Learning, 114(8), 191. https://doi.org/10.1007/s10994-025-06826-w
Download Paper

Deep multi-objective reinforcement learning for utility-based infrastructural maintenance optimization

Published in Neural Computing and Applications, 2025

MO-DCMAC is the first deep multi-objective RL method of its kind applied to a real-world infrastructure problem. Rather than collapsing cost and probability of collapse into a shaped reward, it optimizes directly over non-linear utilities — including FMECA, the risk-scoring methodology Amsterdam’s asset managers already use in practice. It produced the best maintenance plan in 5 of 6 test settings, improving on the rule-based strategies currently in use by up to 46%.

Recommended citation: van Remmerden, J., Kenter, M., Roijers, D. M., Andriotis, C., Zhang, Y., & Bukhsh, Z. (2025). Deep multi-objective reinforcement learning for utility-based infrastructural maintenance optimization. Neural Computing and Applications. https://doi.org/10.1007/s00521-024-10954-0
Download Paper

DROVER: Diffusion-Based Routing with Verbal Preferences

Published in Under review, 2026

A language-conditioned discrete diffusion model for vehicle routing that takes its constraints as natural-language requests, together with the first benchmark and automatic verifier for this setting.

Recommended citation: van Remmerden, J., Wu, Y., Zhang, Y., & Bukhsh, Z. DROVER: Diffusion-Based Routing with Verbal Preferences. Under review.

A Deep Multi-Objective Reinforcement Learning Approach for Infrastructural Maintenance Planning with Non-Linear Utility Functions

Published in Multi-Objective Decision Making Workshop (MODeM) at ECAI 2023, Kraków, 2023

The workshop paper that introduced the multi-objective actor–critic approach to infrastructural maintenance planning later developed into MO-DCMAC. Presented at the Multi-Objective Decision Making (MODeM) workshop at ECAI 2023 in Kraków.

Recommended citation: van Remmerden, J., Kenter, M., Roijers, D. M., Zhang, Y., & Andriotis, C. (2023). A Deep Multi-Objective Reinforcement Learning Approach for Infrastructural Maintenance Planning with Non-Linear Utility Functions. Multi-Objective Decision Making Workshop (MODeM) at ECAI 2023, Kraków, Poland.
Download Paper