Research map / Reinforcement learning / Visual and representation RL
Skill-based Meta-Reinforcement Learning
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 |
|---|---|---|---|
| Combine offline-learned skills and a skill prior with meta-training on task distributions | SiMPL · Maze Navigation | sample efficiency | Helped |
| Combine offline-learned skills and a skill prior with meta-training on task distributions | SiMPL · Kitchen Manipulation | task completion | Helped |
| Use task-agnostic offline data to learn skills and a skill prior | PEARL · Maze Navigation and Kitchen Manipulation | target-task performance | Helped |
| Add meta-training to skill-based RL | SPiRL · Maze Navigation and Kitchen Manipulation | sample efficiency | Helped |
| Use meta-learning rather than multi-task policy distillation to leverage training tasks | SiMPL vs. MTRL · Maze Navigation and Kitchen Manipulation | sample efficiency | Helped |
| Use temporally extended skills instead of flat behavioral cloning from offline data (ablation) | SiMPL vs. BC+PEARL · Maze Navigation, long-horizon meta-training tasks | learning efficiency | Helped |
| Use offline-pretrained behavior cloning with PEARL meta-training (ablation) | BC+PEARL · Maze Navigation, medium-difficulty meta-training tasks | learning efficiency | Helped |
| Use offline-pretrained behavior cloning with MAML meta-training (ablation) | BC+MAML · Maze Navigation, easiest meta-training tasks | environment interactions during meta-training | Hurt |
| Reduce the number of meta-training tasks to 10 or 20 (ablation) | SiMPL · Maze Navigation | sample efficiency versus SPiRL | Helped |
| Train on a meta-training task distribution better aligned with target tasks (ablation) | SiMPL · Maze Navigation | target-task performance | Helped |
| Train on a meta-training task distribution misaligned with target tasks (ablation) | SiMPL · Maze Navigation | target-task performance | Hurt |
| Use the task-agnostic skill prior rather than a uniform prior during meta-training (ablation) | SiMPL · Maze Navigation | target-task performance | Helped |
Closest papers by meaning
- Decoupling Meta-Reinforcement Learning with Gaussian Task Contexts and Skills (AAAI 2024 · similarity 0.93)
- Offline Meta-Reinforcement Learning with Online Self-Supervision (ICML 2022 · similarity 0.93)
- Enhanced Meta Reinforcement Learning via Demonstrations in Sparse Reward Environments (NeurIPS 2022 · similarity 0.92)
- MetaCURE: Meta Reinforcement Learning with Empowerment-Driven Exploration (ICML 2021 · similarity 0.92)
- MetaCARD: Meta-Reinforcement Learning with Task Uncertainty Feedback via Decoupled Context-Aware Reward and Dynamics Components (AAAI 2024 · similarity 0.91)
- Meta-Reinforcement Learning via Exploratory Task Clustering (AAAI 2024 · 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.