Research map / Reinforcement learning / Policy gradient and actor-critic
Actor-critic theory
Analyze convergence, sample efficiency, and implicit bias of actor-critic algorithms.
76 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Policy gradient and actor-critic, keyword cluster “actor critic · policy actor · critic algorithms”
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.
13 papers in 2023–24 and 21 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 3.8% to 4.5%.
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 · The Option-Critic Architecture (AAAI 2017 · 708 citations)
- 2018 · Self-Imitation Learning (ICML 2018 · 140 citations)
- 2019 · Learning to Perform Local Rewriting for Combinatorial Optimization (NeurIPS 2019 · 156 citations)
- 2020 · Model-Augmented Actor-Critic: Backpropagating through Paths (ICLR 2020 · 42 citations)
- 2021 · Single-Timescale Actor-Critic Provably Finds Globally Optimal Policy (ICLR 2021 · 17 citations)
- 2022 · Stackelberg Actor-Critic: Game-Theoretic Reinforcement Learning Algorithms (AAAI 2022 · 38 citations)
- 2023 · Decision-Aware Actor-Critic with Function Approximation and Theoretical Guarantees (NeurIPS 2023 · 1 citation)
- 2024 · Diffusion Actor-Critic with Entropy Regulator (NeurIPS 2024 · 10 citations)
- 2025 · CTD4 – a Deep Continuous Distributional Actor-Critic Agent with a Kalman Fusion of Multiple Critics (AAAI 2025 · 3 citations)
- 2026 · AC-ODM: Actor–Critic Online Data Mixing for Sample-Efficient LLM Pretraining (ICML 2026 · 0 citations)
Earliest papers here
- The Option-Critic Architecture (AAAI 2017 · 708 citations)
- Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation (NeurIPS 2017 · 461 citations)
- An Actor-Critic Algorithm for Sequence Prediction (ICLR 2017 · 215 citations)
What has been tried
54 papers here report 478 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 11%, hurt 17%, 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 |
|---|---|---|---|---|
| Add the entropy regulator to diffusion actor-critic | Walker2d-v3 · return | Helped | Diffusion Actor-Critic with Entropy Regulator (NeurIPS 2024 · 10 citations) | |
| Replace minimum-value or average fusion with Kalman fusion | Cheetah Run and Cartpole Swingup · evaluation reward | Helped | CTD4 – a Deep Continuous Distributional Actor-Critic Agent with a Kalman Fusion of Multiple Critics (AAAI 2025 · 3 citations) | |
| Remove positive-advantage clipping | MuJoCo · performance | Hurt | ReLU to the Rescue: Improve Your On-Policy Actor-Critic with Positive Advantages (ICML 2024 · 2 citations) | |
| add a pessimistic actor alongside the optimistic actor | 30 tasks (RR = 2) · normalized performance | 0.08 -> 1.00 | Helped | Decoupled Policy Actor-Critic: Bridging Pessimism and Risk Awareness in Reinforcement Learning (AAAI 2025 · 1 citation) |
| Remove policy resets and use action sampling for approximate optimism | antmaze-medium-diverse-v2 · performance | Mixed | Actor-Critics Can Achieve Optimal Sample Efficiency (ICML 2025 · 0 citations) | |
| Increase value-improvement gradient steps on the same batch | DeepMind continuous control environments · performance | Helped | Value Improved Actor Critic Algorithms (NeurIPS 2025 · 0 citations) | |
| Use a distributional critic instead of a standard SAC critic | HalfCheetah-v3, Walker2d-v3, and Swimmer-v3 · average return | Helped | Langevin Soft Actor-Critic: Efficient Exploration through Uncertainty-Driven Critic Learning (ICLR 2025 · 0 citations) | |
| replace dynamically updated momentum gradient weights with fixed weights initialized in the first iteration | Kuaishou recommendation logs · Click | 0.539 -> 0.541 | Helped | Finite-Time Convergence and Sample Complexity of Actor-Critic Multi-Objective Reinforcement Learning (ICML 2024 · 0 citations) |
| Use a successor-features critic rather than naive distance-to-skill minimization | All evaluated skill-learning tasks · distance to skill | Helped | Quality-Diversity Actor-Critic: Learning High-Performing and Diverse Behaviors via Value and Successor Features Critics (ICML 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 10 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 · 4 of 10 papers
Baselines
- TD3 · 4 of 10 papers
- SAC · 3 of 10 papers
- REDQ · 2 of 10 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 34 papers; from OpenAlex and the papers' affiliations).
- Google DeepMind · 3 papers
- University of Oxford · 2 papers
- National University of Singapore · 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
- Off-policy actor-critic with expressive policies (71 papers)
- Stabilizing deep actor-critic (48 papers)
Elsewhere in the topic
- Model-based RL with guarantees (245 papers)
- Regularized scalable deep value learning (216 papers)
- Temporal-difference and discounting theory (143 papers)
- Tree search and policy optimization (109 papers)
- Safety layers and shielding (85 papers)
- Variance-reduced policy gradients (79 papers)
- Trajectory reuse in policy gradients (68 papers)
- Distributional value learning (60 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).