Research map / Reinforcement learning / Visual and representation RL
RD-HRL: Generating Reliable Sub-Goals for Long-Horizon Sparse-Reward Tasks
Yixiang Shan, Haipeng Liu, Ting Long, Yi Chang
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 reliability-driven decision mechanism (transition-region targets) to HIQL | HIQL · antmaze-ultra-diverse | normalized score: 52.9 -> 81.1 | Improved |
| add reliability-driven decision mechanism to HIQL | HIQL · CALVIN | normalized score: 43.8 -> 68.8 | Improved |
| use transition regions as decision-level targets instead of state at t+2H (ablation) | RD-HRL · antmaze-ultra-diverse | normalized score: 59.8 -> 81.1 | Improved |
| keep high-level policy between TI module and low-level policy (ablation) | RD-HRL · antmaze-ultra-diverse | normalized score: 27.8 -> 81.1 | Improved |
| use Target Evaluation module instead of plain value V (ablation) | RD-HRL · antmaze-ultra-diverse | normalized score: 35.3 -> 81.1 | Improved |
| add cross-trajectory value update in TE module (ablation) | RD-HRL · antmaze-ultra-diverse | normalized score: 68.1 -> 81.1 | Improved |
| keep temporal abstraction in TE (RD-HRL-CU vs RD-HRL-TE) (ablation) | RD-HRL · antmaze-ultra-diverse | normalized score: 35.3 -> 68.1 | Improved |
| increase waysteps H for HIQL (ablation) | HIQL · antmaze-ultra-play/diverse | normalized score | Mixed |
| use transition-region targets in non-Euclidean manipulation (ablation) | RD-HRL · CALVIN | normalized score: 35.5 -> 68.8 | Improved |
| RD-HRL versus best baseline on kitchen-partial | RD-HRL · kitchen-partial | normalized score: 75.0 -> 69.6 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- HIQL — HIQL: Offline Goal-Conditioned RL with Latent States as Actions (NeurIPS 2023) · better on 5 of 5 shared results
- PlanDQ — PlanDQ: Hierarchical Plan Orchestration via D-Conductor and Q-Performer (ICML 2024) · better on 3 of 4 shared results
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
- Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2026 · similarity 0.88)
- Progress Reward Model for Reinforcement Learning via Large Language Models (NeurIPS 2025 · similarity 0.87)
- Option-aware Temporally Abstracted Value for Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- PlanDQ: Hierarchical Plan Orchestration via D-Conductor and Q-Performer (ICML 2024 · similarity 0.87)
- Learning Process Rewards via Visitation Matching for Efficient RL (NeurIPS 2026 · similarity 0.87)
- Stitching Sub-trajectories with Conditional Diffusion Model for Goal-Conditioned Offline RL (AAAI 2024 · similarity 0.87)
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.
