Research map / Reinforcement learning / Visual and representation RL
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 |
|---|---|---|---|
| Compose short-horizon TDM-25 behavior with planned subgoals | TDM-25 / LEAP · Push and Reach | final puck distance: under 10 cm | Helped |
| Plan over multiple subgoals rather than act greedily toward the final goal | LEAP · Ant Navigation | success | Helped |
| Plan over latent states instead of optimizing subgoals directly in pixel space (ablation) | LEAP · 2D Navigation | task performance | Helped |
| Reuse the pretrained VAE encoder in the RL networks (ablation) | LEAP · 2D Navigation | learning speed | Helped |
| Use the ℓ∞-norm instead of the ℓ1-norm for feasibility (ablation) | LEAP · 2D Navigation | task performance | Helped |
| Use CEM instead of gradient-based optimizers (ablation) | LEAP · 2D Navigation | policy performance | Helped |
| Increase the VAE-prior likelihood penalty weight (ablation) | LEAP · 2D Navigation, Push and Reach, and Ant Navigation | final RL performance | Mixed |
| Use LEAP planning rather than a long-horizon TDM policy | TDM-100 · Push and Reach | environment steps to reach final puck distance under 10 cm: 5 times as many samples | Helped |
| Use short-horizon TDM-25 without planning over subgoals | TDM-25 · Push and Reach | task performance | Hurt |
| Reuse latent-space planning and goal-conditioned reachability estimates | LEAP · 2D Navigation and Push and Reach | task performance | Helped |
| Use model-free methods without multi-stage planning | HER, HER+, HIRO, and TDM baselines · Ant Navigation | success | Hurt |
Closest papers by meaning
- PcLast: Discovering Plannable Continuous Latent States (ICML 2024 · similarity 0.89)
- Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2026 · similarity 0.88)
- Autonomous Option Invention for Continual Hierarchical Reinforcement Learning and Planning (AAAI 2025 · similarity 0.88)
- Interpreting Emergent Planning in Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Reconciling Spatial and Temporal Abstractions for Goal Representation (ICLR 2024 · similarity 0.88)
- Horizon Generalization in Reinforcement Learning (ICLR 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.