Research map / Reinforcement learning / Offline RL
InfraRL: A Benchmark for Constrained Resource Allocation in Large-Scale Infrastructure Asset Management
Yantian Wang, Wenhao Li, Bo Jin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add heuristic imitation loss to CQL | CQL · InfraRL matrix eval | Matrix eval (normalized): 0.4302 -> 0.7082 | Improved |
| add heuristic guidance to CQL | CQL · InfraRL matrix eval | FQE score: -0.1043 -> -0.0911 | Improved |
| switch allocator from score greedy to knapsack (ablation) | CQL · InfraRL | Improve vs History: 0.430 -> 0.703 | Improved |
| switch allocator from score greedy to knapsack (ablation) | CQL-Heuristic · InfraRL | Improve vs History: 0.708 -> 1.000 | Improved |
| switch allocator from score greedy to knapsack (ablation) | MultiTask-BC · InfraRL | Improve vs History: -0.042 -> -0.021 | No clear effect |
| switch allocator from score greedy to knapsack (ablation) | QMIX-CQL-MF · InfraRL | Improve vs History: 0.085 -> 0.444 | Improved |
| use mean-field aggregation in QMIX-CQL | QMIX-CQL · InfraRL | Improve vs History: -0.067 -> 0.085 | Improved |
| use multi-agent QMIX-CQL instead of single-agent CQL | offline RL · InfraRL | Violation rate: 0.007 -> 0.711 | Hurt |
| increase budget factor from 1x to 4x for heuristic-guided value learning (ablation) | CQL-Heuristic · 100-year simulation | final average health: 3.41 -> 3.78 | Improved |
| increase budget factor up to 4.0x for pure imitation (ablation) | Discrete-BC · 100-year simulation | health | No clear effect |
| use value learning (CQL, CPQ) instead of random policy | offline RL · 100-year simulation | final average health: 2.60 -> 3.04 | Improved |
Closest papers by meaning
- Solver-Informed RL: Grounding Large Language Models for Authentic Optimization Modeling (NeurIPS 2025 · similarity 0.83)
- Distilling Reinforcement Learning Algorithms for In-Context Model-Based Planning (ICLR 2025 · similarity 0.82)
- MM-OptBench: A Solver-Grounded Benchmark for Multimodal Optimization Modeling (NeurIPS 2026 · similarity 0.82)
- Autoregressive Policy Optimization for Constrained Allocation Tasks (NeurIPS 2024 · similarity 0.82)
- Constraint-Adaptive Policy Switching for Offline Safe Reinforcement Learning (AAAI 2025 · similarity 0.81)
- Reward Penalties on Augmented States for Solving Richly Constrained RL Effectively (AAAI 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
