Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Invariant Representations for Reinforcement Learning without Reconstruction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use DBC's bisimulation representation instead of DeepMDP | SAC-based driving agent · CARLA Town04 driving | successes (100m): 17% -> 24% | Helped |
| Use DBC's bisimulation representation instead of SAC from raw images | Driving agent · CARLA Town04 driving | distance (m): 123.2\pm7.43 -> 179.0\pm11.4 | Helped |
| Use DBC's bisimulation representation instead of DeepMDP | SAC-based driving agent · CARLA Town04 driving | crash intensity: 1958\pm15.6 -> 2673\pm38.5 | Hurt |
| Use DBC's bisimulation representation instead of DeepMDP | SAC-based driving agent · CARLA Town04 driving | average steer: 10.4\%\pm0.015\% -> 7.3\%\pm0.012\% | Helped |
| Use DBC's bisimulation representation instead of SAC from raw images | Driving agent · CARLA Town04 driving | average brake: 1.3\%\pm0.006\% -> 1.6\%\pm0.022\% | Hurt |
| Replace the bisimulation loss with a reconstruction loss (ablation) | DBC representation learning · DMC natural video distractors | control performance | Hurt |
| Replace the bisimulation loss with contrastive predictive coding (ablation) | DBC representation learning · DMC natural video distractors | control performance | Hurt |
| Train on simple distractors, then transfer without training on natural video backgrounds | DBC representation · DMC natural video distractors | control performance | Helped |
| Reuse the walker-walk representation for a new reward | DBC representation · Walker stand and walker run | control performance | Helped |
| Use the learned bisimulation encoder instead of the prior bisimulation distance encoder | Control policy · DMC control tasks | learning speed | Helped |
Closest papers by meaning
- BeigeMaps: Behavioral Eigenmaps for Reinforcement Learning from Images (ICML 2024 · similarity 0.89)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 2024 · similarity 0.88)
- Learning Fused State Representations for Control from Multi-View Observations (ICML 2025 · similarity 0.88)
- SiT: Symmetry-invariant Transformers for Generalisation in Reinforcement Learning (ICML 2024 · similarity 0.88)
- Learning Latent Dynamic Robust Representations for World Models (ICML 2024 · similarity 0.88)
- State Chrono Representation for Enhancing Generalization in Reinforcement Learning (NeurIPS 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.