Research map / Reinforcement learning / Offline RL
Provably Efficient Offline Goal-Conditioned Reinforcement Learning with General Function Approximation and Single-Policy Concentrability
Hanlin Zhu, Amy Zhang
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 |
|---|---|---|---|
| Tune α from 1 to 0.5 (ablation) | VP-learning / GoFAR · FetchReach | discounted return: 27.3 ± 0.34 -> 27.5 ± 0.29 | No clear effect |
| Tune α from 1 to 0.2 (ablation) | VP-learning / GoFAR · FetchPick | discounted return: 18.2 ± 1.2 -> 20.7 ± 0.9 | Improved |
| Tune α from 1 to 0.1 (ablation) | VP-learning / GoFAR · FetchPush | discounted return: 19.6 ± 1.6 -> 20.3 ± 2.5 | Improved |
| Tune α from 1 to 0.5 (ablation) | VP-learning / GoFAR · FetchSlide | discounted return: 2.75 ± 1.84 -> 3.47 ± 1.08 | Mixed |
| Tune α from 1 to 0.05 (ablation) | VP-learning / GoFAR · HandReach | discounted return: 7.13 ± 3.60 -> 8.98 ± 3.11 | Mixed |
| Use α=0.1 instead of α=1 (ablation) | VP-learning / GoFAR · FetchPush | final distance: 0.043 ± 0.012 -> 0.027 ± 0.002 | Improved |
| Use α=0.2 instead of α=1 (ablation) | VP-learning / GoFAR · FetchPick | final distance: 0.043 ± 0.011 -> 0.031 ± 0.004 | Improved |
| Use α=0.5 instead of α=1 (ablation) | VP-learning / GoFAR · FetchSlide | final distance: 0.1202 ± 0.019 -> 0.1017 ± 0.017 | Improved |
| Use α=0.5 instead of α=1 (ablation) | VP-learning / GoFAR · HandReach | final distance: 0.026828 ± 0.0044 -> 0.026826 ± 0.0049 | No clear effect |
| Use α=2 instead of α=1 (ablation) | VP-learning / GoFAR · FetchSlide | discounted return: 2.75 ± 1.84 -> 1.80 ± 0.66 | Hurt |
| Use α=0.2 instead of α=1 (ablation) | VP-learning / GoFAR · HandReach | discounted return: 7.13 ± 3.60 -> 2.92 ± 0.98 | Hurt |
Closest papers by meaning
- OGBench: Benchmarking Offline Goal-Conditioned RL (ICLR 2025 · similarity 0.92)
- Score Models for Offline Goal-Conditioned Reinforcement Learning (ICLR 2024 · similarity 0.91)
- Stitching Sub-trajectories with Conditional Diffusion Model for Goal-Conditioned Offline RL (AAAI 2024 · similarity 0.91)
- Conservative Offline Goal-Conditioned Implicit V-Learning (ICML 2025 · similarity 0.90)
- Online Optimization for Offline Safe Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.90)
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.
