Research map / Reinforcement learning
Visual and representation RL: research map
2,264 papers on Visual and representation RL in Reinforcement learning, from ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR (2016–2026), grouped into 6 clusters and 23 approaches. The busiest year so far is 2026.
What it is
Practical deep reinforcement learning for agents that see images or act in rich environments, including robots. Key questions are how to learn good state representations and world models, how to explore with sparse rewards, how to reach goals and reuse skills, and how to adapt quickly to new tasks.
Terms you will see on this page:
- World model
- a learned model of how the environment responds, used for planning or imagination.
- Intrinsic motivation
- rewarding the agent for novelty or curiosity when the real reward is sparse.
- Goal-conditioned policy
- a policy told which goal to reach.
Start reading with the most cited papers here:
- A Deep Reinforced Model for Abstractive Summarization (ICLR 2018 · 1,255 citations)
- Benchmarking Deep Reinforcement Learning for Continuous Control (ICML 2016 · 949 citations)
Within Reinforcement learning, its share shrank from 36.5% in 2023–24 to 30.0% in 2025–26 (572 → 696 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Visual and representation RL in the interactive map
Working on something in this topic? Describe your idea in scime atlas to see which approach it falls under, the closest papers by meaning and how crowded the spot has become.
What changed recently
Approaches whose share of the topic moved most between 2023–24 and 2025–26 (ICML, NeurIPS, ICLR, CVPR and AAAI):
- Dexterous manipulation transfer rising: 4% → 14% of the topic (22 → 100 papers)
- RL for combinatorial routing rising: 5% → 8% of the topic (29 → 58 papers)
- Skill learning and diffusion policies rising: 3% → 6% of the topic (17 → 39 papers)
- Latent-space world models falling: 6% → 3% of the topic (32 → 19 papers)
- RL with tree search for discovery falling: 7% → 5% of the topic (41 → 35 papers)
Papers from 2025–2026 that are already cited most:
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · 172 citations)
- Gymnasium: A Standard Interface for Reinforcement Learning Environments (NeurIPS 2025 · 21 citations)
- FlowPolicy: Enabling Fast and Robust 3D Flow-Based Policy via Consistency Flow Matching for Robot Manipulation (AAAI 2025 · 20 citations)
- Pareto Set Learning for Multi-Objective Reinforcement Learning (AAAI 2025 · 14 citations)
Approaches and key papers
RL systems and decision-making · 1,062 papers
Keywords: search, decision, systems
Approaches in this cluster:
- In-context and adaptive RL (205 papers)
Enable fast adaptation in RL via in-context learning, memory, and domain transfer. - Scalable RL systems and environments (188 papers)
Scale RL training and benchmark environments. - Probabilistic and autonomous RL (184 papers)
Frame RL as inference and learn without episodic resets. - RL for combinatorial routing (174 papers)
Solve routing and NP-hard problems with learned RL policies and search. - RL with tree search for discovery (128 papers)
Search molecules and plans with RL and Monte Carlo tree search. - Inferring decision-making from behavior (101 papers)
Interpret human and animal decisions via inverse RL and policy recovery. - Curriculum and environment design (82 papers)
Generates training environments and curricula to improve agent generalization and skill acquisition.
Most cited and most cited since 2024:
- A Deep Reinforced Model for Abstractive Summarization (ICLR 2018 · 1,255 citations)
- Action-Decision Networks for Visual Tracking With Deep Reinforcement Learning (CVPR 2017 · 574 citations)
- Prompt to Transfer: Sim-to-Real Transfer for Traffic Signal Control with Prompt Learning (AAAI 2024 · 41 citations)
- EarnHFT: Efficient Hierarchical Reinforcement Learning for High Frequency Trading (AAAI 2024 · 21 citations)
Representations and world models · 357 papers
Keywords: representation, latent, generalization
Approaches in this cluster:
- Learned state and action representations (150 papers)
Learn compact representations of states, actions, or contexts to speed up and generalize RL. - Robust visual encoders for RL (108 papers)
Make pixel-based RL generalize to visual perturbations via pretrained encoders, invariance, and attention. - Latent-space world models (85 papers)
Learn latent dynamics models and temporal-consistency objectives for model-based and transfer RL. - Successor features for transfer (14 papers)
Use successor representations and features to reuse value information across tasks and policies.
Most cited and most cited since 2024:
- Benchmarking Deep Reinforcement Learning for Continuous Control (ICML 2016 · 949 citations)
- Recurrent World Models Facilitate Policy Evolution (NeurIPS 2018 · 452 citations)
- Jointly Training and Pruning CNNs via Learnable Agent Guidance and Alignment (CVPR 2024 · 20 citations)
- What Effects the Generalization in Visual Reinforcement Learning: Policy Consistency with Truncated Return Prediction (AAAI 2024 · 9 citations)
Robot manipulation · 310 papers
Keywords: robot, manipulation, simulation
Approaches in this cluster:
- Skill learning and diffusion policies (91 papers)
Learn transferable skills and diffusion policies for continuous control. - Dexterous manipulation transfer (164 papers)
Transfer manipulation skills with residual learning, benchmarks, and cross-embodiment data. - Vision-based robot RL (55 papers)
Learn transferable visual control policies and world models for embodied agents.
Most cited and most cited since 2024:
- SAPIEN: A SimulAted Part-Based Interactive ENvironment (CVPR 2020 · 402 citations)
- Hindsight Experience Replay (NeurIPS 2017 · 337 citations)
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · 172 citations)
- Hierarchical Diffusion Policy for Kinematics-Aware Multi-Task Robotic Manipulation (CVPR 2024 · 54 citations)
Exploration and skill discovery · 268 papers
Keywords: exploration, intrinsic, skills
Approaches in this cluster:
- Novelty and impact-driven exploration (87 papers)
Reward visiting novel or impactful states, or diversify behavior, to explore better. - Hierarchical skills and priors (85 papers)
Learn reusable sub-policies and skill priors for hierarchical reinforcement learning. - Reward shaping for sparse rewards (51 papers)
Shape or learn intrinsic rewards and curricula to cope with sparse reward signals. - Unsupervised skill discovery (45 papers)
Discover diverse skills without reward using mutual-information and contrastive objectives.
Most cited and most cited since 2024:
- Hierarchical Deep Reinforcement Learning: Integrating Temporal Abstraction and Intrinsic Motivation (NeurIPS 2016 · 822 citations)
- Parameter Space Noise for Exploration (ICLR 2018 · 359 citations)
- Robust Policy Learning via Offline Skill Diffusion (AAAI 2024 · 6 citations)
- Bootstrapped Reward Shaping (AAAI 2025 · 5 citations)
Goal-conditioned and hierarchical RL · 162 papers
Keywords: goal, conditioned, hierarchical
Approaches in this cluster:
- Goal-conditioned policy learning (70 papers)
Train goal-reaching policies with learned goal representations, discriminative rewards, and pretraining. - Graph search and compositional goals (56 papers)
Plan over goal graphs or compositional goal abstractions to improve goal-conditioned RL. - Subgoal generation for hierarchical RL (36 papers)
Generate adjacency-constrained or probabilistic subgoals so high-level policies guide low-level control.
Most cited and most cited since 2024:
- Visual Reinforcement Learning with Imagined Goals (NeurIPS 2018 · 307 citations)
- Data-Efficient Hierarchical Reinforcement Learning (NeurIPS 2018 · 248 citations)
- Unveiling the Significance of Toddler-Inspired Reward Transition in Goal-Oriented Reinforcement Learning (AAAI 2024 · 3 citations)
- Goal Conditioned Reinforcement Learning for Photo Finishing Tuning (NeurIPS 2024 · 3 citations)
Meta-RL · 105 papers
Keywords: meta, adaptation, context
Approaches in this cluster:
- Task-inference meta-RL (59 papers)
Meta-train context encoders and task selection so agents infer new tasks and adapt quickly. - Task structure and curricula for meta-RL (46 papers)
Exploit subtask structure, curricula, and sampling choices to improve meta-RL adaptation.
Most cited and most cited since 2024:
- Efficient Off-Policy Meta-Reinforcement Learning via Probabilistic Context Variables (ICML 2019 · 214 citations)
- Meta-Reinforcement Learning of Structured Exploration Strategies (NeurIPS 2018 · 174 citations)
- Generalizable Task Representation Learning for Offline Meta-Reinforcement Learning with Data Limitations (AAAI 2024 · 12 citations)
- Meta-Reinforcement Learning via Exploratory Task Clustering (AAAI 2024 · 7 citations)
Related topics in Reinforcement learning
- MDPs and planning (726)
- Policy gradient and actor-critic (1,570)
- Imitation learning (450)
- RL from feedback (758)
- Offline RL (646)
