Research map / Reinforcement learning / Visual and representation RL
Hierarchical and Interpretable Skill Acquisition in Multi-task Reinforcement Learning
Tianmin Shu, Caiming Xiong, Richard Socher
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 |
|---|---|---|---|
| Use the hierarchical policy instead of a flat policy | Minecraft policies trained in small rooms and tested in big rooms · Minecraft big room, Find x | success rate: 0.515 -> 0.723 | Improved |
| Use the hierarchical policy instead of a flat policy | Minecraft policies trained in small rooms and tested in big rooms · Minecraft big room, Get x | success rate: 0.450 -> 0.648 | Improved |
| Train a hierarchical rather than flat policy | Get x policy trained with one item and tested with distracting items · Get x with multiple distracting items | success rate: 29% -> 94% | Improved |
| Add the two-phase curriculum that first trains reuse of base skills (ablation) | Hierarchical policy · Minecraft task learning | convergence speed | Improved |
| Add the stochastic temporal grammar (ablation) | Hierarchical policy · Minecraft task learning | learning efficiency | Improved |
| Add alternating updates of the sub-policies (ablation) | Hierarchical policy · Minecraft task learning | learning efficiency | Improved |
| Use a separate switch value function V_k^sw instead of V_k (ablation) | Hierarchical policy · Minecraft task learning | learning efficiency | Improved |
| Use the hierarchical model rather than the fine-tuned flat policy | Policy π1 trained from the same base policy π0 · G1 learning | average reward: below 0.8 | Improved |
| Use the full hierarchical model rather than a flat policy | Policy π1 trained from the same base policy π0 · G1 learning | episodes to convergence: 22,000 episodes | Improved |
Closest papers by meaning
- Goal-Oriented Skill Abstraction for Offline Multi-Task Reinforcement Learning (ICML 2025 · similarity 0.89)
- Skill Machines: Temporal Logic Skill Composition in Reinforcement Learning (ICLR 2024 · similarity 0.89)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.88)
- Policy Compatible Skill Incremental Learning via Lazy Learning Interface (NeurIPS 2025 · similarity 0.88)
- When Do Skills Help Reinforcement Learning? A Theoretical Analysis of Temporal Abstractions (ICML 2024 · similarity 0.88)
- Skill-SD: Skill-Conditioned Self-Distillation for Multi-turn LLM Agents (NeurIPS 2026 · 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.
