Research map / Reinforcement learning / Visual and representation RL
Bisimulation Makes Analogies in Goal-Conditioned Reinforcement Learning
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 goal-conditioned bisimulation representations with analogy-based goal specification | IQL · Analogy | success rate: 0.280 ± 0.024 -> 0.403 ± 0.041 | Helped |
| Use goal-conditioned bisimulation representations | IQL · Drawer | success rate: 0.463 ± 0.023 -> 0.533 ± 0.037 | Helped |
| Use goal-conditioned bisimulation representations | IQL · Drawer+VD | success rate: 0.283 ± 0.030 -> 0.448 ± 0.033 | Helped |
| Use goal-conditioned bisimulation representations | IQL · BD | success rate: 0.522 ± 0.042 -> 0.440 ± 0.026 | Hurt |
| Use goal-conditioned bisimulation representations | IQL · BD+VD | success rate: 0.225 ± 0.028 -> 0.322 ± 0.021 | Helped |
| Use goal-conditioned bisimulation representations | IQL · Analogy+VD | success rate: 0.145 ± 0.029 -> 0.303 ± 0.027 | Helped |
| Add goal-point grounding φ(g,g) to the ψ objective (ablation) | GCB · Button and Drawer | success rate | Helped |
| Use an ℓ1 rather than ℓ2 metric loss for φ (ablation) | GCB · Button and Drawer | success rate | Helped |
| Backpropagate critic gradients into the encoder (ablation) | GCB · Button and Drawer | success rate | Hurt |
| Remove the reward decoder (ablation) | GCB · Button and Drawer | success rate | No clear effect |
| Use sampled next observations rather than a learned dynamics model (ablation) | GCB · Button and Drawer | success rate | Helped |
| Vary the latent dimensionality of φ and ψ (ablation) | GCB · Drawer | success rate | No clear effect |
Closest papers by meaning
- Hierarchical Goal Abstractions via Learned Subset Relations (ICML 2026 · similarity 0.90)
- Discrete Compositional Representations as an Abstraction for Goal Conditioned Reinforcement Learning (NeurIPS 2022 · similarity 0.90)
- Building Minimal and Reusable Causal State Abstractions for Reinforcement Learning (AAAI 2024 · similarity 0.89)
- Reconciling Spatial and Temporal Abstractions for Goal Representation (ICLR 2024 · similarity 0.87)
- A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals (ICLR 2025 · similarity 0.86)
- Improving the Continuity of Goal-Achievement Ability via Policy Self-Regularization for Goal-Conditioned Reinforcement Learning (ICML 2025 · 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.