Research map / Reinforcement learning / Visual and representation RL
RL for combinatorial routing
Solve routing and NP-hard problems with learned RL policies and search.
158 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Visual and representation RL, keyword cluster “search · decision · systems”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
29 papers in 2023–24 and 58 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 5.1% to 8.3%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Efficient Macroscopic Urban Traffic Models for Reducing Congestion: A PDDL+ Planning Approach (AAAI 2016 · 84 citations)
- 2017 · Action-Decision Networks for Visual Tracking With Deep Reinforcement Learning (CVPR 2017 · 574 citations)
- 2018 · Cellular Network Traffic Scheduling With Deep Reinforcement Learning (AAAI 2018 · 224 citations)
- 2019 · A Deep Reinforcement Learning Framework for Rebalancing Dockless Bike Sharing Systems (AAAI 2019 · 140 citations)
- 2020 · Toward A Thousand Lights: Decentralized Deep Reinforcement Learning for Large-Scale Traffic Signal Control (AAAI 2020 · 419 citations)
- 2021 · Online 3D Bin Packing with Constrained Deep Reinforcement Learning (AAAI 2021 · 140 citations)
- 2022 · CADRE: A Cascade Deep Reinforcement Learning Framework for Vision-Based Autonomous Urban Driving (AAAI 2022 · 38 citations)
- 2023 · SafeLight: A Reinforcement Learning Method toward Collision-Free Traffic Signal Control (AAAI 2023 · 49 citations)
- 2024 · Prompt to Transfer: Sim-to-Real Transfer for Traffic Signal Control with Prompt Learning (AAAI 2024 · 41 citations)
- 2025 · CarPlanner: Consistent Auto-regressive Trajectory Planning for Large-Scale Reinforcement Learning in Autonomous Driving (CVPR 2025 · 14 citations)
- 2026 · WorldRFT: Latent World Model Planning with Reinforcement Fine-Tuning for Autonomous Driving (AAAI 2026 · 5 citations)
Earliest papers here
- Efficient Macroscopic Urban Traffic Models for Reducing Congestion: A PDDL+ Planning Approach (AAAI 2016 · 84 citations)
- Strategic Attentive Writer for Learning Macro-Actions (NeurIPS 2016 · 80 citations)
- Generating Long-term Trajectories Using Deep Hierarchical Networks (NeurIPS 2016 · 72 citations)
What has been tried
113 papers here report 1,121 measured changes (a component added, removed or swapped, compared with a baseline): helped 68%, no clear effect 8%, hurt 20%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add quality reward alongside DE reward | Test14-Random non-reactive · S-CR | 97.49 -> 99.22 | Helped | CarPlanner: Consistent Auto-regressive Trajectory Planning for Large-Scale Reinforcement Learning in Autonomous Driving (CVPR 2025 · 14 citations) |
| Add the trajectory prediction module | CARLA training · collision avoidance | Helped | Graph-Based Prediction and Planning Policy Network (GP3Net) for Scalable Self-Driving in Dynamic Environments Using Deep Reinforcement Learning (AAAI 2024 · 8 citations) | |
| Use Top-2 routing rather than Hard Routing or Soft Routing | Meta-World · success rate | Helped | Not All Tasks Are Equally Difficult: Multi-Task Deep Reinforcement Learning with Dynamic Depth Routing (AAAI 2024 · 5 citations) | |
| add RL post-training to an IL policy | 3DGS evaluation benchmark · CR | 0.229 -> 0.089 | Helped | RAD: Training an End-to-End Driving Policy via Large-Scale 3DGS-based Reinforcement Learning (NeurIPS 2025 · 4 citations) |
| Multiview graph (micro+meso+macro, diameters 2+5+10 km) vs single-view graphs | Manhattan/Hangzhou ablation · scheduling error | Helped | GARLIC: GPT-Augmented Reinforcement Learning with Intelligent Control for Vehicle Dispatching (AAAI 2025 · 3 citations) | |
| Remove LRBS from fine-tuning | TSP200 · optimality gap | 0.50% -> 0.52% | Hurt | Scaling Combinatorial Optimization Neural Improvement Heuristics with Online Search and Adaptation (AAAI 2025 · 3 citations) |
| Add work-conserving probability masking to PPO | Queuing benchmark instances · pair-wise win rate against traditional policies | 77% | Helped | QGym: Scalable Simulation and Benchmarking of Queuing Network Controllers (NeurIPS 2024 · 3 citations) |
| add trace-based pruning | CartPole, 1 initial state · runtime (seconds) | 3.13 -> 0.22 | Helped | In Search of Trees: Decision-Tree Policy Synthesis for Black-Box Systems via Search (AAAI 2025 · 2 citations) |
| Add Agent Phase to the deliberative reasoning branch | Songdo · AEWT | 9.03 -> 8.12 | Helped | VLMLight: Safety-Critical Traffic Signal Control via Vision-Language Meta-Control and Dual-Branch Reasoning Architecture (NeurIPS 2025 · 2 citations) |
| add action-free video pre-training | nuScenes planning · Avg. L2 (m) | 3.34 -> 2.29 | Helped | From Forecasting to Planning: Policy World Model for Collaborative State-Action Prediction (NeurIPS 2025 · 2 citations) |
| use multiple samples and voting for plan selection instead of single sampling | GridMaze2D 15×15 (vr=2) · success rate | 0.060±0.022 -> 0.114±0.025 | Helped | Versatile Navigation Under Partial Observability via Value-guided Diffusion Policy (CVPR 2024 · 2 citations) |
| Add simulated annealing to DNC | Joint inventory replenishment · test return | Helped | Dynamic Neighborhood Construction for Structured Large Discrete Action Spaces (ICLR 2024 · 2 citations) |
What reviewers will expect
Datasets and baselines named in the 28 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- nuScenes · 3 of 28 papers
Baselines
- PPO · 3 of 28 papers
- POMO · 2 of 28 papers
- SGBS · 2 of 28 papers
- EAS · 2 of 28 papers
- Concorde · 2 of 28 papers
- IDM · 2 of 28 papers
- Vanilla PPO · 2 of 28 papers
- UniAD · 2 of 28 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 87 papers; from OpenAlex and the papers' affiliations).
- Huawei · 6 papers
- Tianjin University · 4 papers
- Hong Kong University of Science and Technology · 4 papers
- University of Science and Technology of China · 3 papers
- Tsinghua University · 3 papers
- Jilin University · 3 papers
- Carnegie Mellon University · 3 papers
- Stanford University · 3 papers
- Shanghai Jiao Tong University · 3 papers
- Nanyang Technological University · 3 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- In-context and adaptive RL (202 papers)
- Scalable RL systems and environments (179 papers)
- Probabilistic and autonomous RL (177 papers)
- RL with tree search for discovery (121 papers)
- Inferring decision-making from behavior (93 papers)
- Curriculum and environment design (81 papers)
Elsewhere in the topic
- Learned state and action representations (148 papers)
- Dexterous manipulation transfer (141 papers)
- Skill learning and diffusion policies (86 papers)
- Robust visual encoders for RL (86 papers)
- Novelty and impact-driven exploration (85 papers)
- Hierarchical skills and priors (84 papers)
- Latent-space world models (83 papers)
- Goal-conditioned policy learning (66 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).