Research map / Reinforcement learning / Visual and representation RL
Hierarchical skills and priors
Learn reusable sub-policies and skill priors for hierarchical reinforcement learning.
84 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Visual and representation RL, keyword cluster “exploration · intrinsic · skills”
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.
27 papers in 2023–24 and 21 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 4.7% to 3.0%.
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 · Hierarchical Deep Reinforcement Learning: Integrating Temporal Abstraction and Intrinsic Motivation (NeurIPS 2016 · 822 citations)
- 2017 · Modular Multitask Reinforcement Learning with Policy Sketches (ICML 2017 · 214 citations)
- 2018 · Parameter Space Noise for Exploration (ICLR 2018 · 359 citations)
- 2019 · Hierarchical Reinforcement Learning with Advantage-Based Auxiliary Rewards (NeurIPS 2019 · 35 citations)
- 2020 · Reinforcement Learning with Perturbed Rewards (AAAI 2020 · 102 citations)
- 2021 · PEBBLE: Feedback-Efficient Interactive Reinforcement Learning via Relabeling Experience and Unsupervised Pre-training (ICML 2021 · 13 citations)
- 2022 · Causality-driven Hierarchical Structure Discovery for Reinforcement Learning (NeurIPS 2022 · 13 citations)
- 2023 · Incremental Reinforcement Learning with Dual-Adaptive ε-Greedy Exploration (AAAI 2023 · 10 citations)
- 2024 · Robust Policy Learning via Offline Skill Diffusion (AAAI 2024 · 6 citations)
- 2025 · Enhancing Question Generation through Diversity-Seeking Reinforcement Learning with Bilevel Policy Decomposition (AAAI 2025 · 3 citations)
- 2026 · Learn the Ropes, Then Trust the Wins: Self-imitation with Progressive Exploration for Agentic Reinforcement Learning (ICLR 2026 · 1 citation)
Earliest papers here
- Hierarchical Deep Reinforcement Learning: Integrating Temporal Abstraction and Intrinsic Motivation (NeurIPS 2016 · 822 citations)
- Sparse Latent Space Policy Search (AAAI 2016 · 10 citations)
- Modular Multitask Reinforcement Learning with Policy Sketches (ICML 2017 · 214 citations)
What has been tried
67 papers here report 693 measured changes (a component added, removed or swapped, compared with a baseline): helped 60%, no clear effect 11%, hurt 22%, mixed 7%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add hierarchical embeddings and guided diffusion decoder | Multi-stage Meta-World Speed, Energy, and Wind Level 3 · achieved reward | 3.59 ± 0.19 -> 3.81 ± 0.06; 3.16 ± 0.35 -> 3.67 ± 0.14; 3.19 ± 0.17 -> 3.44 ± 0.18 | Helped | Robust Policy Learning via Offline Skill Diffusion (AAAI 2024 · 6 citations) |
| Add bilevel policy decomposition to SFT | SQuAD 1.1 / 1 · Top-1 BLEU-4 | 18.52 -> 19.41 | Helped | Enhancing Question Generation through Diversity-Seeking Reinforcement Learning with Bilevel Policy Decomposition (AAAI 2025 · 3 citations) |
| remove measure functions | Humanoid · Best Reward | 9677 -> 9653 | No clear effect | Proximal Policy Gradient Arborescence for Quality Diversity Reinforcement Learning (ICLR 2024 · 3 citations) |
| add NGM | Nguyen benchmarks, 10 noisy inputs · RR | 32.9% -> 89.1% | Helped | Noise-Resilient Symbolic Regression with Dynamic Gating Reinforcement Learning (AAAI 2025 · 2 citations) |
| Replace DGPO's pairwise diversity measurement with mutual information | MPE Spread (hard) · diversity score | Hurt | DGPO: Discovering Multiple Strategies with Diversity-Guided Policy Optimization (AAAI 2024 · 2 citations) | |
| Use automatic curriculum learning by sampling preferred context regions for each expert | Table Tennis · success rate | Helped | Acquiring Diverse Skills using Curriculum Reinforcement Learning with Mixture of Experts (ICML 2024 · 2 citations) | |
| Increase r_min range from 0.1 to 0.5 | Coverage · performance | Helped | Efficient Model-Based Reinforcement Learning Through Optimistic Thompson Sampling (ICLR 2025 · 1 citation) | |
| replace state-dependent Q-switch selection with uniform policy selection | Multistage Reacher · success rate | 60% | Hurt | QMP: Q-switch Mixture of Policies for Multi-Task Behavior Sharing (ICLR 2025 · 1 citation) |
| EMA smoothing instead of Gaussian | RoboDesk, Hand, Earthmoving · task performance | Hurt | DreamSmooth: Improving Model-based Reinforcement Learning via Reward Smoothing (ICLR 2024 · 1 citation) | |
| Apply the unbiased action-bias correction objective | discrete Bayesian optimization · search efficiency | Helped | In-context Exploration-Exploitation for Reinforcement Learning (ICLR 2024 · 1 citation) | |
| Add outcome teleportation to OC-GFN pre-training | GridWorld, especially larger maps · sample efficiency | Helped | Pre-Training and Fine-Tuning Generative Flow Networks (ICLR 2024 · 1 citation) | |
| remove the epistemic-uncertainty intrinsic reward | DMC state-based tasks · episodic return | Hurt | SOMBRL: Scalable and Optimistic Model-Based RL (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 16 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- DeepMind Control Suite · 2 of 16 papers
Baselines
- SAC · 3 of 16 papers
- PPO · 2 of 16 papers
- HUCRL · 2 of 16 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 48 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 4 papers
- University of Montreal · 4 papers
- Montreal Institute for Learning Algorithms · 3 papers
- Carnegie Mellon University · 2 papers
- Microsoft · 2 papers
- Sungkyunkwan University · 2 papers
- Stanford University · 2 papers
- Google DeepMind · 2 papers
- Université de Montréal · 2 papers
- University of Alberta · 2 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
- Novelty and impact-driven exploration (85 papers)
- Reward shaping for sparse rewards (51 papers)
- Unsupervised skill discovery (45 papers)
Elsewhere in the topic
- In-context and adaptive RL (202 papers)
- Scalable RL systems and environments (179 papers)
- Probabilistic and autonomous RL (177 papers)
- RL for combinatorial routing (158 papers)
- Learned state and action representations (148 papers)
- Dexterous manipulation transfer (141 papers)
- RL with tree search for discovery (121 papers)
- Inferring decision-making from behavior (93 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).