Research map / Reinforcement learning / Visual and representation RL
Skill learning and diffusion policies
Learn transferable skills and diffusion policies for continuous control.
86 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Visual and representation RL, keyword cluster “robot · manipulation · simulation”
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.
17 papers in 2023–24 and 39 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 3.0% to 5.6%.
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 · Hindsight Experience Replay (NeurIPS 2017 · 337 citations)
- 2018 · Learning an Embedding Space for Transferable Robot Skills (ICLR 2018 · 175 citations)
- 2019 · Policy Transfer with Strategy Optimization (ICLR 2019 · 56 citations)
- 2020 · RL-CycleGAN: Reinforcement Learning Aware Simulation-to-Real (CVPR 2020 · 158 citations)
- 2021 · Self-Supervised Policy Adaptation during Deployment (ICLR 2021 · 60 citations)
- 2022 · Accelerated Policy Learning with Parallel Differentiable Simulation (ICLR 2022 · 15 citations)
- 2023 · Policy Adaptation From Foundation Model Feedback (CVPR 2023 · 4 citations)
- 2024 · SPOC: Imitating Shortest Paths in Simulation Enables Effective Navigation and Manipulation in the Real World (CVPR 2024 · 26 citations)
- 2025 · Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · 172 citations)
- 2026 · From Prior to Pro: Efficient Skill Mastery via Distribution Contractive RL Finetuning (ICML 2026 · 0 citations)
Earliest papers here
- Hindsight Experience Replay (NeurIPS 2017 · 337 citations)
- Grounded Action Transformation for Robot Learning in Simulation (AAAI 2017 · 78 citations)
- Generalizing Skills with Semi-Supervised Reinforcement Learning (ICLR 2017 · 34 citations)
What has been tried
59 papers here report 588 measured changes (a component added, removed or swapped, compared with a baseline): helped 64%, no clear effect 9%, hurt 21%, 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 |
|---|---|---|---|---|
| replace the Transformer action decoder with a GRU | CHORES-S · average success | 49.9 -> 43.6 | Hurt | SPOC: Imitating Shortest Paths in Simulation Enables Effective Navigation and Manipulation in the Real World (CVPR 2024 · 26 citations) |
| distill the soft decision tree | Kitchen MKBL · ACS | 2.90 ± 0.42 -> 3.06 ± 0.65 | Helped | SkillTree: Explainable Skill-Based Deep Reinforcement Learning for Long-Horizon Control Tasks (AAAI 2025 · 4 citations) |
| Add the policy-filter gate | MetaWorld-MT10 · success rate | Helped | Efficient Multi-task Reinforcement Learning with Cross-Task Policy Guidance (NeurIPS 2024 · 4 citations) | |
| Add trajectory length mixing | Tabular random MDP · average reward | Helped | Differentiable Information Enhanced Model-Based Reinforcement Learning (AAAI 2025 · 2 citations) | |
| Set the success-rate target to α=0.5 rather than higher values | Hopper, Walker2D, and HalfCheetah · global success rate | Helped | Domain Randomization via Entropy Maximization (ICLR 2024 · 2 citations) | |
| search the inheritance ratio with SC-BO instead of fixing it at 0% | Ant · MTS | 10.25±1.31 -> 12.07±0.60 | Helped | Efficient Language-instructed Skill Acquisition via Reward-Policy Co-Evolution (AAAI 2025 · 1 citation) |
| add cross-domain action consistency loss to the skill adapter | Multi-stage Metaworld, wind domain · success rate | 5.4% to 6.3% | Helped | In-Context Policy Adaptation via Cross-Domain Skill Diffusion (AAAI 2025 · 1 citation) |
| add task-oriented perturbation instead of random perturbation | Hopper Hop, Disassemble, Coffee Push, and Hammer · normalized sample efficiency (training time) | Hopper Hop: 1 -> 0.6167; Disassemble: 0.8505 -> 0.7056; Coffee Push: 0.9481 -> 0.8066; Hammer: 0.875 -> 0.7167 | Helped | MENTOR: Mixture-of-Experts Network with Task-Oriented Perturbation for Visual Reinforcement Learning (ICML 2025 · 1 citation) |
| Train the dynamics model with noised actions | CALVIN ArticulatedParts · steering success | Helped | DynaGuide: Steering Diffusion Polices with Active Dynamic Guidance (NeurIPS 2025 · 1 citation) | |
| Set source-distribution std to 0.1 instead of 0.2 | LIBERO average · success rate | 94.8 -> 77.5 | Hurt | Conditioning Matters: Training Diffusion Policies is Faster Than You Think (NeurIPS 2025 · 1 citation) |
| Set shielding threshold to v_thr=0 instead of v_thr=-0.05 or v_thr=-0.10 | Vanilla-Env Lab training · safety violations | Hurt | Sim-to-Lab-to-Real: Safe Reinforcement Learning with Shielding and Generalization Guarantees (Abstract Reprint) (AAAI 2024 · 1 citation) | |
| Add loss guidance | MetaWorld, language context · CR | 65.58 ± 5.02 -> 90.38 ± 3.14 | Helped | LLM-based Skill Diffusion for Zero-shot Policy Adaptation (NeurIPS 2024 · 1 citation) |
What reviewers will expect
Datasets and baselines named in the 22 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MetaWorld · 3 of 22 papers
- DeepMind Control Suite · 2 of 22 papers
- Adroit · 2 of 22 papers
- LIBERO · 2 of 22 papers
Baselines
- SAC · 2 of 22 papers
- DreamerV3 · 2 of 22 papers
- SPiRL · 2 of 22 papers
- DrQ-v2 · 2 of 22 papers
- pi0 · 2 of 22 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 56 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 6 papers
- Nanjing University · 4 papers
- Amazon · 4 papers
- University of Maryland, College Park · 4 papers
- Meta · 3 papers
- Carnegie Mellon University · 3 papers
- Stanford University · 3 papers
- Massachusetts Institute of Technology · 3 papers
- Chinese Academy of Sciences · 2 papers
- Peking University · 2 papers
Use it
Download BibTeX of 38 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
- Dexterous manipulation transfer (141 papers)
- Vision-based robot RL (44 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)
- RL with tree search for discovery (121 papers)
- Inferring decision-making from behavior (93 papers)
- Robust visual encoders for RL (86 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).