Research map / Reinforcement learning / Visual and representation RL
Goal-Conditioned Reinforcement Learning with Imagined Subgoals
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 imagined subgoals to define a prior policy and KL-regularize policy improvement | RIS · Ant navigation mazes | success rate | Helped |
| Use imagined subgoals to define a prior policy and KL-regularize policy improvement | RIS · Vision-based robotic manipulation | success rate | Helped |
| Use imagined subgoals during training, then use a flat policy without test-time subgoal planning | RIS · S-shaped and ω-shaped ant mazes | success rate | Helped |
| Replace the imagined-subgoal prior with a uniform action prior (ablation) | RIS · Ant U-maze | success rate | Hurt |
| Replace the imagined-subgoal prior with an exponential-moving-average policy prior (ablation) | RIS · Ant U-maze | success rate | Hurt |
| Remove implicit regularization that keeps predicted subgoals near the valid state distribution (ablation) | RIS high-level policy · Ant U-maze | success rate | Hurt |
| Use oracle subgoals instead of subgoals predicted by the learned high-level policy (ablation) | RIS · Ant U-maze | learning speed | Helped |
| Train TDM with a long policy horizon and dense distance-to-goal rewards | TDM · Ant navigation mazes | learning | Hurt |
| Plan and sequentially reach subgoals at inference instead of using them only to guide training | LEAP · Ant navigation mazes | sample efficiency | Hurt |
| Increase task horizon | SAC and LEAP · S-shaped and ω-shaped ant mazes | success rate | Hurt |
Closest papers by meaning
- Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2026 · similarity 0.90)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.89)
- Hierarchical Goal Abstractions via Learned Subset Relations (ICML 2026 · similarity 0.89)
- Discrete Compositional Representations as an Abstraction for Goal Conditioned Reinforcement Learning (NeurIPS 2022 · similarity 0.89)
- Probabilistic Subgoal Representations for Hierarchical Reinforcement Learning (ICML 2024 · similarity 0.88)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.88)
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.