Research map / Reinforcement learning / Visual and representation RL
KAGE-Bench: Fast Known-Axis Visual Generalization Evaluation for Reinforcement Learning
Egor Cherepanov, Daniil Zelezetsky, Aleksandr Panov, Alexey Kovalev
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| shift eval filters (photometric) vs train config | PPO-CNN · KAGE-Bench Filters | success rate: 0.83 -> 0.11 | Hurt |
| shift eval lighting effects vs train config | PPO-CNN · KAGE-Bench Effects | success rate: 0.82 -> 0.16 | Hurt |
| shift eval background vs train config | PPO-CNN · KAGE-Bench Background | success rate: 0.90 -> 0.42 | Hurt |
| shift agent appearance at eval | PPO-CNN · KAGE-Bench Agent | SR gap (%): 21.1% | Hurt |
| shift distractors at eval | PPO-CNN · KAGE-Bench Distractors | SR gap (%): 30.9% | Hurt |
| train on more visual diversity (3 images bg vs one image bg) | PPO-CNN · KAGE-Bench Background | SR gap (%): 9.6 -> -1.3 | Improved |
| add RAD-PPO cutout-color augmentation | PPO-CNN · KAGE-Bench 8 hardest configs | success rate | Mixed |
| add crop-resize augmentation, CURL, DrQ, SVEA | PPO-CNN · KAGE-Bench hard configs | convergence | Hurt |
| increase same-as-agent distractors from 0 to 11 at eval | PPO-CNN · KAGE-Bench Distractors | success rate | Hurt |
Closest papers by meaning
- VIGOR: Zero-Shot Visual Generalization via Latent-Space Consistency in Model-Based Reinforcement Learning (NeurIPS 2026 · similarity 0.89)
- Focus On What Matters: Separated Models For Visual-Based RL Generalization (NeurIPS 2024 · similarity 0.89)
- Sliding Puzzles Gym: A Scalable Benchmark for State Representation in Visual Reinforcement Learning (ICML 2025 · similarity 0.88)
- The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
- SiT: Symmetry-invariant Transformers for Generalisation in Reinforcement Learning (ICML 2024 · similarity 0.88)
- DrM: Mastering Visual Reinforcement Learning through Dormant Ratio Minimization (ICLR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
