Research map / Reinforcement learning / Visual and representation RL
Learning Domain Invariant Representations in Goal-conditioned Block MDPs
Beining Han, Chongyi Zheng, Harris Chan, Keiran Paster, Michael R. Zhang, Jimmy Ba
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 |
|---|---|---|---|
| Add perfect alignment losses (MMD and difference) with aligned sampling to Skew-Fit | Skew-Fit · Reach test environments | hand distance: 0.111 -> 0.076 | Improved |
| Add perfect alignment to Skew-Fit | Skew-Fit · Door test environments | angle difference: 0.194 -> 0.106 | Improved |
| Add perfect alignment to Skew-Fit | Skew-Fit · Push test environments | puck distance: 0.086 -> 0.069 | Improved |
| Add perfect alignment to Skew-Fit | Skew-Fit · Pickup test environments | object distance: 0.037 -> 0.028 | Improved |
| Add RAD data augmentation to Skew-Fit | Skew-Fit · Pickup test environments | object distance: 0.037 -> 0.040 | Hurt |
| Add RAD data augmentation to Skew-Fit | Skew-Fit · Door test environments | angle difference: 0.194 -> 0.162 | Improved |
| Use bisimulation methods (MISA, DBC) instead of PA-SF | Skew-Fit · Door test environments | angle difference: 0.106 -> 0.255 | Hurt |
| Train PA-SF on training environments vs oracle single-environment Skew-Fit | PA-SF · training environments | distance metrics: 0.058 -> 0.057 | No clear effect |
| Remove difference loss L_DIFF (PA-SF w/o D) (ablation) | PA-SF · Door | latent error rate and test performance | Hurt |
| Remove both MMD and difference losses (PA-SF w/o MD) (ablation) | PA-SF · Door | latent error rate | Hurt |
| Remove aligned sampling (PA-SF w/o AS) (ablation) | PA-SF · Door training environments | training performance | Hurt |
| Add MMD loss for alignment (ablation) | PA-SF · Door | latent error rate on training environments | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Skew-Fit — Skew-Fit: State-Covering Self-Supervised Reinforcement Learning (ICML 2020) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DSAP: Enhancing Generalization in Goal-Conditioned Reinforcement Learning (AAAI 2026 · similarity 0.89)
- Horizon Generalization in Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.88)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.87)
- Towards General-Purpose Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.87)
- What Effects the Generalization in Visual Reinforcement Learning: Policy Consistency with Truncated Return Prediction (AAAI 2024 · similarity 0.86)
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.
