Research map / Reinforcement learning / Visual and representation RL
Hierarchical Reinforcement Learning for Zero-shot Generalization with Subtask Dependencies
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 |
|---|---|---|---|
| Use NSGS instead of GRProp | Subtask graph execution · Playground D1-D4 zero-shot | normalized reward: .721 -> .820; .682 -> .785; .623 -> .715; .424 -> .527 | Helped |
| Use GRProp instead of Greedy | Subtask graph execution · Playground D1-D4 zero-shot | normalized reward: .164 -> .721; .144 -> .682; .178 -> .623; .228 -> .424 | Helped |
| Use NSGS instead of Greedy | Subtask graph execution · Mining Eval zero-shot | mean reward: 3.39 -> 8.19 | Helped |
| Use GRProp instead of Greedy | Subtask graph execution · Mining Eval zero-shot | mean reward: 3.39 -> 6.16 | Helped |
| Use NSGS instead of GRProp | Subtask graph execution · Mining Eval zero-shot | mean reward: 6.16 -> 8.19 | Helped |
| Use NSGS instead of Independent | Subtask graph execution · Playground D1-D4 adaptation | normalized reward: .346 -> .828; .296 -> .797; .193 -> .733; .188 -> .552 | Helped |
| Use NSGS instead of Independent | Subtask graph execution · Mining Eval adaptation | mean reward: 3.89 -> 8.58 | Helped |
| Use NSGS rather than MCTS alone as the planning policy | MCTS on Playground · Playground D2 | normalized reward: approximately 0.97 normalized reward with 33K simulations | Helped |
| Add actor-critic fine-tuning after GRProp distillation | NSGS · Mining and Playground training | reward | Helped |
| Use observation-based cost scores alongside graph reward scores | NSGS · Mining domain | reward | Helped |
Closest papers by meaning
- Zero-Shot Task Generalization with Multi-Task Deep Reinforcement Learning (ICML 2017 · similarity 0.91)
- Robust Subtask Learning for Compositional Generalization (ICML 2023 · similarity 0.90)
- Meta Reinforcement Learning with Autonomous Inference of Subtask Dependencies (ICLR 2020 · similarity 0.90)
- One Subgoal at a Time: Zero-Shot Generalization to Arbitrary Linear Temporal Logic Requirements in Multi-Task Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Zero Shot Generalization of Vision-Based RL Without Data Augmentation (ICML 2025 · similarity 0.88)
- Foundation Policies with Hilbert Representations (ICML 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.