Research map / Reinforcement learning / Visual and representation RL
Flexible Attention-Based Multi-Policy Fusion for Efficient Deep Reinforcement Learning
Zih-Yun Chiu, Yi-Lin Tuan, William Yang Wang, Michael C. Yip
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 |
|---|---|---|---|
| KIAN vs KoGuN when transferring Pick-and-Place 1x to 10x goal range | KIAN · Zero-shot S2C Pick-and-Place 10x | success rate: 0.24 -> 0.72 | Improved |
| KIAN vs KoGuN on Empty-Random 5x5 to 16x16 | KIAN · Zero-shot S2C Empty-Random 16x16 | success rate: 0.53 -> 0.93 | Improved |
| KIAN vs KoGuN on DoorKey-5x5 to 8x8 | KIAN · Zero-shot S2C DoorKey 8x8 | success rate: 0.77 -> 0.76 | No clear effect |
| KIAN vs KoGuN on Slide 1x to 5x | KIAN · Zero-shot S2C Slide 5x | success rate: 0.55 -> 0.42 | Hurt |
| KIAN vs KoGuN on Slide 1x to 10x | KIAN · Zero-shot S2C Slide 10x | success rate: 0.18 -> 0.15 | Hurt |
| KIAN vs KoGuN on Push 1x to 5x | KIAN · Zero-shot S2C Push 5x | success rate: 0.63 -> 0.93 | Improved |
| KIAN vs KoGuN on Pick-and-Place to Reach | KIAN · Zero-shot C2S Reach | success rate: 0.45 -> 1.00 | Improved |
| KIAN vs RL on Pick-and-Place to Push | KIAN · Zero-shot C2S Push | success rate: 0.31 -> 0.30 | No clear effect |
| KIAN vs KoGuN on DoorKey-8x8 to Unlock | KIAN · Zero-shot C2S Unlock | success rate: 0.91 -> 0.94 | Improved |
| A2T applied to continuous-control transfer instead of KIAN | A2T · Zero-shot S2C Slide 5x | success rate: 0.42 -> 0.00 | Hurt |
| Do not address entropy imbalance (remove modified policy distributions) in KIAN (ablation) | KIAN · MiniGrid and OpenAI-Robotics learning curves | sample efficiency | Improved |
| Entropy imbalance in continuous control, FetchPickAndPlace and FetchPush (ablation) | KIAN without fix · FetchPickAndPlace, FetchPush | success rate | Hurt |
Closest papers by meaning
- Self-Supervised Attention-Aware Reinforcement Learning (AAAI 2021 · similarity 0.86)
- Learning Agentic Policy from Action Guidance (NeurIPS 2026 · similarity 0.85)
- Mixture-of-Experts Meets In-Context Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- Unveiling Markov heads in Pretrained Language Models for Offline Reinforcement Learning (ICML 2025 · similarity 0.85)
- Towards an Information Theoretic Framework of Context-Based Offline Meta-Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
- NoRD: A Data-Efficient Vision-Language-Action Model that Drives without Reasoning (CVPR 2026 · similarity 0.85)
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.
