Research map / Reinforcement learning / Visual and representation RL
Reward shaping for sparse rewards
Shape or learn intrinsic rewards and curricula to cope with sparse reward signals.
51 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · 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.
11 papers in 2023–24 and 17 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 1.9% to 2.4%.
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 · A Laplacian Framework for Option Discovery in Reinforcement Learning (ICML 2017 · 73 citations)
- 2018 · Learning by Playing Solving Sparse Reward Tasks from Scratch (ICML 2018 · 255 citations)
- 2019 · Large-Scale Study of Curiosity-Driven Learning (ICLR 2019 · 356 citations)
- 2020 · Learning to Utilize Shaping Rewards: A New Approach of Reward Shaping (NeurIPS 2020 · 99 citations)
- 2021 · Learning Task-Distribution Reward Shaping with Meta-Learning (AAAI 2021 · 14 citations)
- 2022 · Reinforcement Learning with Stochastic Reward Machines (AAAI 2022 · 16 citations)
- 2023 · Learning to Shape Rewards Using a Game of Two Partners (AAAI 2023 · 4 citations)
- 2024 · Potential-Based Reward Shaping for Intrinsic Motivation (Student Abstract) (AAAI 2024 · 2 citations)
- 2025 · Bootstrapped Reward Shaping (AAAI 2025 · 5 citations)
- 2026 · Hierarchical Reinforcement Learning for Sparse-Reward Search in Commutative Algebra (ICML 2026 · 0 citations)
Earliest papers here
- A Laplacian Framework for Option Discovery in Reinforcement Learning (ICML 2017 · 73 citations)
- Learning by Playing Solving Sparse Reward Tasks from Scratch (ICML 2018 · 255 citations)
- Belief Reward Shaping in Reinforcement Learning (AAAI 2018 · 73 citations)
What has been tried
42 papers here report 438 measured changes (a component added, removed or swapped, compared with a baseline): helped 56%, no clear effect 9%, hurt 29%, 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 |
|---|---|---|---|---|
| Tune η to an intermediate value around 2 | Atari suite · median human-normalized score | η≈2 | Mixed | Bootstrapped Reward Shaping (AAAI 2025 · 5 citations) |
| Remove mean adjustment (normalization) | MiniGrid DoorKey 8x8 (α = 0.02, γ = 0.995) · time to convergence | 1.67E6 -> N/A | Hurt | Potential-Based Reward Shaping for Intrinsic Motivation (Student Abstract) (AAAI 2024 · 2 citations) |
| Set alpha to 0.25 rather than 0.5 | 2DMaze, four task returns · episodic return | 0.893 ± 0.033 -> 0.889 ± 0.031; 0.908 ± 0.022 -> 0.901 ± 0.025; 0.924 ± 0.020 -> 0.915 ± 0.023; | Mixed | Centralized Reward Agent for Knowledge Sharing and Transfer in Multi-Task Reinforcement Learning (NeurIPS 2025 · 1 citation) |
| Replace state-only reward with a state-action-pair reward | RobotReach and RobotPush · average episodic return | 170.18±0.00 -> 151.95±5.74 (RobotReach); 167.14±0.00 -> 179.76±1.66 (RobotPush) | Mixed | Highly Efficient Self-Adaptive Reward Shaping for Reinforcement Learning (ICLR 2025 · 1 citation) |
| normalize GRM rewards | Montezuma's Revenge (ALE) · average cumulative extrinsic reward | p=0.37 | No clear effect | Action-Dependent Optimality-Preserving Reward Shaping (ICML 2025 · 0 citations) |
| Remove the relevance term, using only achievability and novelty | pointmaze 2D–6D · steps to 10% target goal achievement (M) | 2.9 vs 5.2; 3.1 vs ∞; 7.4 vs ∞; 5.4 vs ∞; 18.7 vs ∞ | Hurt | DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · 0 citations) |
| Use D³RB rather than other selection algorithms | Continuous control tasks · average normalized task reward | Helped | ORSO: Accelerating Reward Design via Online Reward Selection and Policy Optimization (ICLR 2025 · 0 citations) | |
| Use the stochastic max-reward formulation instead of deterministic max-reward TD3 | Stochastic single-goal Maze, DSP reward k=3 · success ratio | Helped | To the Max: Reinventing Reward in Reinforcement Learning (ICML 2024 · 0 citations) | |
| add goal-reaching and linear rewards to MLP rewards | AntMaze total · normalized score | 25.3 ± 8 -> 47.3 ± 7 | Helped | Unsupervised Zero-Shot Reinforcement Learning via Functional Reward Encodings (ICML 2024 · 0 citations) |
| Add the contribution reward to novelty-only DuRND | sparse-reward tasks · episode return | Helped | Catching Two Birds with One Stone: Reward Shaping with Dual Random Networks for Balancing Exploration and Exploitation (ICML 2025) | |
| Include the policy agent's action in the reward agent input, rather than using state alone | continuous control tasks · episodic return | Helped | Reward Shaping for Reinforcement Learning with An Assistant Reward Agent (ICML 2024) |
What reviewers will expect
Datasets and baselines named in the 12 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MountainCar · 2 of 12 papers
Baselines
- RND · 4 of 12 papers
- ReLara · 3 of 12 papers
- ExploRS · 2 of 12 papers
- #Explo · 2 of 12 papers
- ROSA · 2 of 12 papers
- SORS · 2 of 12 papers
- PPO · 2 of 12 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 28 papers; from OpenAlex and the papers' affiliations).
- University of California, Berkeley · 4 papers
- Google · 3 papers
- National University of Singapore · 3 papers
- University College London · 2 papers
- North Carolina State University · 2 papers
- Nanyang Technological University · 2 papers
- Max Planck Institute for Intelligent Systems · 2 papers
- ETHZ - ETH Zurich · 2 papers
- Google DeepMind · 2 papers
Use it
Download BibTeX of 39 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)
- Hierarchical skills and priors (84 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).