Dejavu: Towards Experience Feedback Learning for Embodied Intelligence
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866917398766944256 |
|---|---|
| author | Wu, Shaokai Ji, Yanbiao Li, Qiuchang Zhang, Zhiyi He, Qichen Xie, Wenyuan Zhang, Guodong Bayramli, Bayram Ding, Yue Lu, Hongtao |
| author_facet | Wu, Shaokai Ji, Yanbiao Li, Qiuchang Zhang, Zhiyi He, Qichen Xie, Wenyuan Zhang, Guodong Bayramli, Bayram Ding, Yue Lu, Hongtao |
| contents | Embodied agents face a fundamental limitation: once deployed in real-world environments, they cannot easily acquire new knowledge to improve task performance. In this paper, we propose Dejavu, a general post-deployment learning framework that augments a frozen Vision-Language-Action (VLA) policy with retrieved execution memories through an Experience Feedback Network (EFN). EFN identifies contextually relevant prior action experiences and conditions action prediction on the retrieved guidance. We train EFN with reinforcement learning and semantic similarity rewards, encouraging the predicted actions to align with past behaviors under the current observation. During deployment, EFN continually expands its memory with new trajectories, enabling the agent to exhibit ``learning from experience.'' Experiments across diverse embodied tasks show that EFN improves adaptability, robustness, and success rates over frozen baselines. Our Project Page is https://dejavu2025.github.io/. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_10181 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Dejavu: Towards Experience Feedback Learning for Embodied Intelligence Wu, Shaokai Ji, Yanbiao Li, Qiuchang Zhang, Zhiyi He, Qichen Xie, Wenyuan Zhang, Guodong Bayramli, Bayram Ding, Yue Lu, Hongtao Robotics Artificial Intelligence Computer Vision and Pattern Recognition Embodied agents face a fundamental limitation: once deployed in real-world environments, they cannot easily acquire new knowledge to improve task performance. In this paper, we propose Dejavu, a general post-deployment learning framework that augments a frozen Vision-Language-Action (VLA) policy with retrieved execution memories through an Experience Feedback Network (EFN). EFN identifies contextually relevant prior action experiences and conditions action prediction on the retrieved guidance. We train EFN with reinforcement learning and semantic similarity rewards, encouraging the predicted actions to align with past behaviors under the current observation. During deployment, EFN continually expands its memory with new trajectories, enabling the agent to exhibit ``learning from experience.'' Experiments across diverse embodied tasks show that EFN improves adaptability, robustness, and success rates over frozen baselines. Our Project Page is https://dejavu2025.github.io/. |
| title | Dejavu: Towards Experience Feedback Learning for Embodied Intelligence |
| topic | Robotics Artificial Intelligence Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2510.10181 |