DROVER: Diffusion-Based Routing with Verbal Preferences
van Remmerden, J., Wu, Y., Zhang, Y., & Bukhsh, Z. DROVER: Diffusion-Based Routing with Verbal Preferences. Under review.
Curriculum Vitae
I am a fourth-year PhD candidate at Eindhoven University of Technology working on reinforcement learning, discrete diffusion models and language models for combinatorial optimization. I expect to defend in January 2027 and am interested in postdoctoral and research positions.
Email: j.v.remmerden@tue.nl · jessevanremmerden@gmail.com
Links: Google Scholar · GitHub · ORCID · LinkedIn · TU/e profile
Education
- PhD Candidate, Artificial Intelligence for Planning and Scheduling, Eindhoven University of Technology (TU/e), Jan. 2023 – Jan. 2027 (expected)
- Information Systems group, Department of Industrial Engineering & Innovation Sciences (IE&IS); AI for Decision Making cluster
- Supervised by Dr. Yingqian Zhang (promotor) and Dr. Zaharah A. Bukhsh (daily supervisor)
- Five first-author papers (three journal, one workshop, one under review); three open-source codebases
- MSc Artificial Intelligence, Utrecht University, Feb. 2020 – Jun. 2022
- Cum Laude, 8.2/10
- Thesis (8.6/10): Utilising Reinforcement Learning for the Diversified Top-k Clique Search Problem
- Honours Young Innovators programme, Utrecht Honours College
- BSc Computer Science, University of Amsterdam, Sept. 2016 – Oct. 2019
- Thesis (9/10): detecting dolphins in video footage with deep learning (Faster R-CNN)
Research
DROVER — routing from instructions written in plain English (2025 – present)
First author; under review. Language-conditioned discrete diffusion. Built the first benchmark for instruction-following vehicle routing (26+ request types, ~3 × 10⁹ phrasings, reference solutions and an automatic verifier), and a diffusion model that reads the request through a frozen language model and steers its solution toward satisfying it. Obeys the stated request in 98–99% of cases on 50–1,000 stops versus ~32% for the strongest existing neural solvers, and 87–90% on request types never seen in training, where existing solvers achieve 0%. Details
CDQAC & Offline-LD — learning factory schedules from data, not simulators (2023 – 2026)
First author; Machine Learning (2025) and TMLR (2026). Offline RL for job shop and flexible job shop scheduling. Offline-LD was the first method to learn JSSP dispatching policies purely from a fixed set of past schedules, matching or beating simulator-trained baselines from only 100 examples. CDQAC went further: trained on schedules produced by random decisions, it matches or beats the best simulator-trained methods across four benchmark suites, scales to problems with ten times as many jobs as seen in training, and needs 10–25 training instances. Established that offline performance is governed by how much of the decision space the data covers rather than the quality of its solutions (rank correlation −0.95). CDQAC · CDQAC project page · Offline-LD
MO-DCMAC — 50-year maintenance planning for Amsterdam’s historic canal walls (2023 – 2025)
First author; Neural Computing and Applications (2025), with the City of Amsterdam. Multi-objective RL. Turned FMECA — the risk-scoring methodology Amsterdam’s asset managers use in practice — into the training objective, so the learned plan optimizes cost and probability of collapse directly. Best plan in 5 of 6 test settings, up to 46% better than the rule-based strategies in use; the first deep multi-objective RL method of its kind applied to a real-world problem. Details
Funded projects
- STABILITY — Sustainable Circular Life Extension Strategies for Inner-City Bridges and Quay Walls (NWA.1431.20.004), 2023 – 2027. Dutch Research Agenda consortium funded by NWO and the City of Amsterdam. Project page
Publications
Generalizing Beyond Suboptimality: Offline Reinforcement Learning Learns Effective Scheduling through Random Solutions
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.
Offline Reinforcement Learning for Learning to Dispatch for Job Shop Scheduling
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
Deep multi-objective reinforcement learning for utility-based infrastructural maintenance optimization
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
A Deep Multi-Objective Reinforcement Learning Approach for Infrastructural Maintenance Planning with Non-Linear Utility Functions
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.
Talks and tutorials
DROVER: Diffusion-Based Routing with Verbal Preferences
Talk at Data Science Meets Optimisation (DSO) Workshop at IJCAI-ECAI 2026, Bremen, Germany
Tutorial: Deep Reinforcement Learning for Combinatorial Optimization at IJCAI-ECAI 2026
Tutorial at IJCAI-ECAI 2026, Bremen, Germany
Generalizing Beyond Suboptimality: Offline RL Learns Effective Scheduling Through Random Solutions
Talk at NWO ICT.OPEN 2026, Smart Industry track, De Doelen, Rotterdam, The Netherlands
Offline Reinforcement Learning for Learning to Dispatch for Job Shop Scheduling at ECML PKDD 2025
Talk at ECML PKDD 2025, Journal Track, Porto, Portugal
Poster Session during DLRL Summer School 2024
Poster at DLRL Summer School 2024, Toronto, Canada
Oral presentation during ICT Open 2024
Talk at NWO ICT.OPEN 2024, Utrecht, The Netherlands
Workshop presentation during MODeM 2023
Talk at Multi-Objective Decision Making Workshop (MODeM) at ECAI 2023, Kraków, Poland
Professional experience
- Blockchain Engineer, Kryha, Amsterdam, Aug. 2018 – Jan. 2020
- Built proofs-of-concept with smart contracts, tokenization, cryptography and federated learning
- First prize, Odyssey Hackathon 2019 (Nature 2.0 track): a federated-learning, blockchain and evolutionary-computing proof-of-concept built in one weekend
Service and leadership
- Member, PhD Council, Eindhoven University of Technology, 2024 – 2025
- Tutorial speaker, Deep Reinforcement Learning for Combinatorial Optimization, IJCAI-ECAI 2026, Bremen
- Co-author of the 2022 municipal election programme, Volt Amsterdam (2021)
- Head of the Health, Welfare and Sports Committee, Jonge Democraten Amsterdam (2016 – 2017)
Technical skills
- Programming: Python (8+ years; PyTorch, PyTorch Lightning, JAX), C/C++, Rust, SQL, LaTeX
- Machine learning: Offline and online reinforcement learning (CQL, QR-DQN, SAC, multi-objective RL), discrete diffusion and classifier-free guidance, graph transformers, conditioning on frozen LLM encoders, reward/verifier/benchmark design, synthetic data generation
- Optimization: OR-Tools (CP-SAT), Gurobi, LKH
- Infrastructure: Linux/HPC (SLURM, Apptainer, Kubernetes/Run:AI), multi-GPU training on A100/H100/B200, Docker, Git, Weights & Biases
- Languages: Dutch (native), English (fluent, C1)
References
- Dr. Yingqian Zhang — PhD promotor, Information Systems, Eindhoven University of Technology — yqzhang@tue.nl
- Dr. Zaharah A. Bukhsh — daily supervisor, Information Systems, Eindhoven University of Technology — z.bukhsh@tue.nl