Research map / Reinforcement learning / Visual and representation RL
Probabilistic and autonomous RL
Frame RL as inference and learn without episodic resets.
177 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · 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.
59 papers in 2023–24 and 61 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 10.3% to 8.8%.
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
- 2017 · Robust Adversarial Reinforcement Learning (ICML 2017 · 366 citations)
- 2018 · Deep Reinforcement Learning That Matters (AAAI 2018 · 1,592 citations)
- 2019 · Learning Reward Machines for Partially Observable Reinforcement Learning (NeurIPS 2019 · 67 citations)
- 2020 · Model Based Reinforcement Learning for Atari (ICLR 2020 · 401 citations)
- 2021 · DEAR: Deep Reinforcement Learning for Online Advertising Impression in Recommender Systems (AAAI 2021 · 115 citations)
- 2022 · How Private Is Your RL Policy? An Inverse RL Based Analysis Framework (AAAI 2022 · 11 citations)
- 2023 · Safe Reinforcement Learning via Shielding under Partial Observability (AAAI 2023 · 43 citations)
- 2024 · Reinforcement Learning Policy as Macro Regulator Rather than Macro Placer (NeurIPS 2024 · 5 citations)
- 2025 · Gymnasium: A Standard Interface for Reinforcement Learning Environments (NeurIPS 2025 · 21 citations)
- 2026 · Phase-Aware Mixture of Experts for Agentic Reinforcement Learning (ICML 2026 · 0 citations)
Earliest papers here
- Robust Adversarial Reinforcement Learning (ICML 2017 · 366 citations)
- Deep Reinforcement Learning That Matters (AAAI 2018 · 1,592 citations)
- Deep Q-learning From Demonstrations (AAAI 2018 · 843 citations)
What has been tried
140 papers here report 1,325 measured changes (a component added, removed or swapped, compared with a baseline): helped 63%, no clear effect 9%, hurt 22%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Remove parameter fusion and use only hypernetwork-generated parameters | MO-MuJoCo · HV | 5.36 vs 1.47 (MO-Walker2d-v2); 5.92 vs 4.65 (MO-HalfCheetah-v2); 8.63 vs 1.39 (MO-Ant-v2); 3.22 vs 2.11 (MO-Swimmer-v2); | Hurt | Pareto Set Learning for Multi-Objective Reinforcement Learning (AAAI 2025 · 14 citations) |
| replace MaskPlace's placer formulation with regulator refinement | ICCAD 2015 superblue1 · Global HPWL | 7.93 ± 0.06 -> 5.58 ± 0.05 | Helped | Reinforcement Learning Policy as Macro Regulator Rather than Macro Placer (NeurIPS 2024 · 5 citations) |
| increase the soft-constraint strength γ | 5Zone base and extreme weather environments · reward | Helped | Active Reinforcement Learning for Robust Building Control (AAAI 2024 · 4 citations) | |
| Use all recorded features for subgroup partitioning instead of feature taxonomy distillation | Probability ITS · student performance | Hurt | Get a Head Start: On-Demand Pedagogical Policy Selection in Intelligent Tutoring (AAAI 2024 · 4 citations) | |
| Change the acquisition function optimism representation (EI versus UCB) | Contextual RL benchmarks · performance | No clear effect | Model-Based Transfer Learning for Contextual Reinforcement Learning (NeurIPS 2024 · 4 citations) | |
| add SNR regularization to PUN | CartPoleSwingUp · ATP | 534.43 ± 48.40 -> 670.48 ± 13.42 | Helped | Regularized Parameter Uncertainty for Improving Generalization in Reinforcement Learning (CVPR 2024 · 3 citations) |
| Increase the CVAE ensemble size beyond 5 | OOD detection across environments · false positive rate | Helped | Guaranteeing Out-Of-Distribution Detection in Deep RL via Transition Estimation (AAAI 2025 · 2 citations) | |
| add word normalization to action prompt probabilities | Tomato Salad · success rate | 0.67 ± 0.47 -> \mathbf{1 ± 0} | No clear effect | True Knowledge Comes from Practice: Aligning Large Language Models with Embodied Environments via Reinforcement Learning (ICLR 2024 · 2 citations) |
| fine-tune the perception module with reward signals instead of freezing it | Atari tasks · task return | Helped | End-to-End Neuro-Symbolic Reinforcement Learning with Textual Explanations (ICML 2024 · 2 citations) | |
| add the critic agent to the slow and fast agents | MineDojo: crafting table · success rate | 0.52 -> 0.65 | Helped | RL-GPT: Integrating Reinforcement Learning and Code-as-policy (NeurIPS 2024 · 2 citations) |
| Use only the decoder head rather than no heads | Dev10 · DS | 17.4 -> 83.5 | Helped | Raw2Drive: Reinforcement Learning with Aligned World Models for End-to-End Autonomous Driving (in CARLA v2) (NeurIPS 2025 · 1 citation) |
| replace the factorized policy with a full covariance policy | Hopper Jump · max jump height | Helped | Open the Black Box: Step-based Policy Updates for Temporally-Correlated Episodic Reinforcement Learning (ICLR 2024 · 1 citation) |
What reviewers will expect
Datasets and baselines named in the 39 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MO-Hopper · 2 of 39 papers
- CartPole · 2 of 39 papers
Baselines
- SAC · 6 of 39 papers
- PPO · 4 of 39 papers
- L2 regularization · 2 of 39 papers
- PG-MORL · 2 of 39 papers
- PD-MORL · 2 of 39 papers
- CAPQL · 2 of 39 papers
- GPI-LS · 2 of 39 papers
- LTL2Action · 2 of 39 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 120 papers; from OpenAlex and the papers' affiliations).
- Carnegie Mellon University · 6 papers
- Nanjing University · 5 papers
- Meta · 5 papers
- University of California, Berkeley · 4 papers
- Peking University · 4 papers
- University of Montreal · 4 papers
- University of Alberta · 4 papers
- Tsinghua University · 3 papers
- University of Cambridge · 3 papers
- University of Oxford · 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)
- RL for combinatorial routing (158 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).