Research map / Reinforcement learning / Visual and representation RL
HIQL: Offline Goal-Conditioned RL with Latent States as Actions
Seohong Park, Dibya Ghosh, Benjamin Eysenbach, Sergey Levine
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 |
|---|---|---|---|
| remove latent representation learning (ablation) | HIQL · kitchen-partial | normalized return: 65.0 -> 46.3 | Hurt |
| remove latent representation learning (ablation) | HIQL · kitchen-mixed | normalized return: 67.7 -> 36.8 | Hurt |
| remove latent representation learning (ablation) | HIQL · calvin | normalized return: 43.8 -> 23.4 | Hurt |
| remove latent representation learning (ablation) | HIQL · antmaze-ultra-play | normalized return: 39.2 vs 56.0 | Mixed |
| remove latent representation learning (ablation) | HIQL · antmaze-large-diverse | normalized return: 88.2 vs 87.3 | No clear effect |
| replace flat policy with hierarchical policy extraction | GC-IQL · antmaze-large-diverse | normalized return: 50.7 -> 88.2 | Improved |
| replace flat policy with hierarchical policy extraction | GC-IQL · procgen-maze-1000-test | success rate: 60.0 -> 78.2 | Improved |
| use temporal abstraction and representation instead of GC-POR hierarchy | GC-POR · calvin | normalized return: 12.4 -> 43.8 | Improved |
| train with 25% action labels (rest action-free) (ablation) | HIQL · antmaze-large-diverse | normalized return: 88.2 -> 88.9 | No clear effect |
| train with 25% action labels (rest action-free) (ablation) | HIQL · antmaze-ultra-diverse | normalized return: 52.9 -> 38.2 | Hurt |
| train with 25% action labels (rest action-free) (ablation) | HIQL · calvin | normalized return: 43.8 -> 35.8 | Hurt |
| train with 25% action labels (rest action-free) (ablation) | HIQL · procgen-maze-500-test | normalized return: 64.5 -> 65.5 | No clear effect |
Later papers that compared with this method
26 later accepted papers report HIQL as a baseline in their results tables; 20 of them report a better number on most of the results they share with it.
- Laplacian Representations for Decision-Time Planning (ICML 2026 · better on 10 of 13 shared results)
- Chain-of-Goals Hierarchical Policy for Long-Horizon Offline Goal-Conditioned RL (ICML 2026 · better on 9 of 10 shared results)
- SVL: Goal-Conditioned Reinforcement Learning as Survival Learning (ICML 2026 · better on 8 of 10 shared results)
- Compositional Transduction with Latent Analogies for Offline Goal-Conditioned Reinforcement Learning (ICML 2026 · better on 7 of 8 shared results)
- Bottleneck-Guided Spectral Subgoals For Offline Goal-Conditioned RL (ICML 2026 · better on 8 of 8 shared results)
- Latent Representation Alignment for Offline Goal-Conditioned Reinforcement Learning (ICML 2026 · better on 8 of 8 shared results)
- Goal Reaching with Eikonal-Constrained Hierarchical Quasimetric Reinforcement Learning (ICLR 2026 · better on 4 of 6 shared results)
- RD-HRL: Generating Reliable Sub-Goals for Long-Horizon Sparse-Reward Tasks (ICLR 2026 · better on 5 of 5 shared results)
18 more not listed.
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.92)
- Stitching Sub-trajectories with Conditional Diffusion Model for Goal-Conditioned Offline RL (AAAI 2024 · similarity 0.91)
- Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2026 · similarity 0.90)
- Option-aware Temporally Abstracted Value for Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.90)
- Are Expressive Models Truly Necessary for Offline RL? (AAAI 2025 · similarity 0.89)
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.
