Research map / Reinforcement learning / Visual and representation RL
Imitating Graph-Based Planning with Goal-Conditioned Policies
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 |
|---|---|---|---|
| Add PIG self-imitation and subgoal skipping | MSS · Large U-shaped AntMaze | success rate: 19.08% -> 57.41% | Helped |
| Add planning-guided self-imitation | MSS · Long-horizon continuous control tasks | sample-efficiency | Helped |
| Train with PIG and remove the planner at test time (ablation) | MSS + PIG · U-shaped AntMaze | success rate | Helped |
| Replace PIG self-imitation with hindsight action imitation (ablation) | MSS + PIG without subgoal skipping · 2DReach, Pusher, L-shaped AntMaze | sample-efficiency | Hurt |
| Add subgoal skipping (ablation) | PIG · 2DReach and Pusher | success rate | Helped |
| Add subgoal skipping (ablation) | PIG · L-shaped AntMaze | success rate | No clear effect |
| Add subgoal skipping (ablation) | PIG · Reacher and Large U-shaped AntMaze | success rate | Mixed |
| Use random subgoal selection instead of loss-conditioned skipping (ablation) | PIG · 2DReach | success rate | No clear effect |
| Use random subgoal selection instead of loss-conditioned skipping (ablation) | PIG · Pusher | success rate | Hurt |
| Set the self-imitation loss coefficient to 1e-3 or 1e-4 rather than 0 (ablation) | PIG · L-shaped AntMaze | success rate | Helped |
| Use too large a self-imitation loss coefficient (ablation) | PIG · L-shaped AntMaze | success rate | Hurt |
| Use subgoal skipping (ablation) | PIG · 2DReach | state entropy | Helped |
Closest papers by meaning
- Learning to Reach Goals via Iterated Supervised Learning (ICLR 2021 · similarity 0.91)
- Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2026 · similarity 0.90)
- Learning to Reach Goals via Diffusion (ICML 2024 · similarity 0.88)
- Breadth-First Exploration on Adaptive Grid for Reinforcement Learning (ICML 2024 · similarity 0.88)
- Stitching Sub-trajectories with Conditional Diffusion Model for Goal-Conditioned Offline RL (AAAI 2024 · similarity 0.88)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · 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.