Latent Bridge: Feature Delta Prediction for Efficient Dual-System Vision-Language-Action Model Inference
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866918481526521856 |
|---|---|
| author | Liu, Yudong Li, Yuan Tang, Zijia Zheng, Yuxi Lin, Yueqian Wang, Qinsi Li, Yi Liu, Shuangjun Zhang, Shuai Jing, Taotao Gao, Dashan Bi, Ning Sun, Jingwei Chen, Yiran Li, Hai |
| author_facet | Liu, Yudong Li, Yuan Tang, Zijia Zheng, Yuxi Lin, Yueqian Wang, Qinsi Li, Yi Liu, Shuangjun Zhang, Shuai Jing, Taotao Gao, Dashan Bi, Ning Sun, Jingwei Chen, Yiran Li, Hai |
| contents | Dual-system Vision-Language-Action (VLA) models achieve state-of-the-art robotic manipulation but are bottlenecked by the VLM backbone, which must
execute at every control step while producing temporally redundant features. We propose Latent Bridge, a lightweight model that predicts VLM output
deltas between timesteps, enabling the action head to operate on predicted outputs while the expensive VLM backbone is called only periodically. We
instantiate Latent Bridge on two architecturally distinct VLAs: GR00T-N1.6 (feature-space bridge) and π0.5 (KV-cache bridge), demonstrating that the
approach generalizes across VLA designs. Our task-agnostic DAgger training pipeline transfers across benchmarks without modification. Across four
LIBERO suites, 24 RoboCasa kitchen tasks, and the ALOHA sim transfer-cube task, Latent Bridge achieves 95-100% performance retention while reducing
VLM calls by 50-75%, yielding 1.65-1.73x net per-episode speedup. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_02739 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Latent Bridge: Feature Delta Prediction for Efficient Dual-System Vision-Language-Action Model Inference Liu, Yudong Li, Yuan Tang, Zijia Zheng, Yuxi Lin, Yueqian Wang, Qinsi Li, Yi Liu, Shuangjun Zhang, Shuai Jing, Taotao Gao, Dashan Bi, Ning Sun, Jingwei Chen, Yiran Li, Hai Robotics Dual-system Vision-Language-Action (VLA) models achieve state-of-the-art robotic manipulation but are bottlenecked by the VLM backbone, which must execute at every control step while producing temporally redundant features. We propose Latent Bridge, a lightweight model that predicts VLM output deltas between timesteps, enabling the action head to operate on predicted outputs while the expensive VLM backbone is called only periodically. We instantiate Latent Bridge on two architecturally distinct VLAs: GR00T-N1.6 (feature-space bridge) and π0.5 (KV-cache bridge), demonstrating that the approach generalizes across VLA designs. Our task-agnostic DAgger training pipeline transfers across benchmarks without modification. Across four LIBERO suites, 24 RoboCasa kitchen tasks, and the ALOHA sim transfer-cube task, Latent Bridge achieves 95-100% performance retention while reducing VLM calls by 50-75%, yielding 1.65-1.73x net per-episode speedup. |
| title | Latent Bridge: Feature Delta Prediction for Efficient Dual-System Vision-Language-Action Model Inference |
| topic | Robotics |
| url | https://arxiv.org/abs/2605.02739 |