Research map / Reinforcement learning / Visual and representation RL
Goal-conditioned Offline Planning from Curious Exploration
Marco Bagatella, Georg Martius
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 model-based planning (horizon H>1) on top of a TD3 actor (ablation) | TD3 value function · maze_large, kitchen, fetch_push, pinpad | success rate | Improved |
| add graph-based value aggregation to model-based planning (ablation) | TD3+MBP · maze_large, pinpad | success rate | Improved |
| model-based planning alone without aggregation on long-horizon tasks (ablation) | TD3+MBP · maze_large, pinpad | success rate | No clear effect |
| sample graph vertices uniformly instead of by inverse density (ablation) | Graph-based value aggregation · long-horizon tasks | success rate | Hurt |
| replace value aggregation with SORB-like subgoal selection (ablation) | Graph-based value aggregation · maze_large, pinpad | success rate | Hurt |
| add model-based planning to IQL, LAPO and WGCSL actors | IQL, LAPO, WGCSL · maze_large | success rate | Improved |
| replace TD3 with IQL, LAPO or WGCSL for value learning | Value learning on curious exploration data · maze_large | success rate | No clear effect |
| reduce dataset from 200k to 100k and 50k transitions (ablation) | TD3, TD3+MBP, TD3+MBP+Aggregation · all four environments | success rate | Hurt |
| reduce graph size from 1000 to 100 vertices | Graph-based value aggregation · pinpad | success rate | No clear effect |
Closest papers by meaning
- Option-aware Temporally Abstracted Value for Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2026 · similarity 0.89)
- Learning World Models for Unconstrained Goal Navigation (NeurIPS 2024 · similarity 0.88)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.88)
- Conservative Offline Goal-Conditioned Implicit V-Learning (ICML 2025 · similarity 0.88)
- Physics-informed Value Learner for Offline Goal-Conditioned Reinforcement Learning (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.
