Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Dejavu: Towards Experience Feedback Learning for Embodied Intelligence
Shaokai Wu, Yanbiao Ji, Qiuchang Li, Zhiyi Zhang, Qichen He, Wenyuan Xie, Guodong Zhang, Bayram Bayramli, Yue Ding, Hongtao Lu
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 EFN with a larger experience bank (Vol=1000 vs Vol=300) (ablation) | OpenVLA + EFN · LIBERO average | success rate (%): 84.4 -> 87.0 | Improved |
| add EFN residual retrieval module to frozen VLA | OpenVLA · LIBERO average | success rate (%): 76.5 -> 87.0 | Improved |
| add EFN to frozen VLA | OpenVLA · LIBERO-Long | success rate (%): 53.7 -> 76.5 | Improved |
| add EFN to stronger base policy | UniVLA · LIBERO average | success rate (%): 95.2 -> 97.2 | Improved |
| replace EFN with kNN-RAG (copy action from nearest past step) | OpenVLA · LIBERO average | success rate (%): 76.5 -> 73.2 | Hurt |
| add retrieval-only kNN-RAG on real robot | GO-1 · AgiBot-G1 real world | success rate (%): 25.8 -> 18.2 | Hurt |
| add test-time finetuning GC-TTT on backbone | GO-1 · AgiBot-G1 real world | success rate (%): 25.8 -> 26.2 | No clear effect |
| add residual-only correction without retrieval (ResAct) | OpenVLA · LIBERO average | success rate (%): 76.5 -> 79.2 | Improved |
| use EFN with frozen backbone instead of R2A backbone updates | GO-1 · AgiBot-G1 real world | success rate (%): 39.5 -> 50.8 | Improved |
| add SAC (entropy-regularized RL) for EFN training (ablation) | OpenVLA + EFN · LIBERO | success rate and steps | Improved |
| add similarity-shaped dense rewards (ablation) | OpenVLA + EFN · LIBERO | success rate | Improved |
| add instruction-level retrieval filtering (ablation) | OpenVLA + EFN · LIBERO | success rate | Improved |
Closest papers by meaning
- How Foundational Skills Influence VLM-based Embodied Agents: A Native Perspective (AAAI 2026 · similarity 0.89)
- Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- VLM Agents Generate Their Own Memories: Distilling Experience into Embodied Programs of Thought (NeurIPS 2024 · similarity 0.87)
- NeSyC: A Neuro-symbolic Continual Learner For Complex Embodied Tasks In Open Domains (ICLR 2025 · similarity 0.86)
- ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning (NeurIPS 2025 · similarity 0.86)
- Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.86)
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.
