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 {\pi}0.5 (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.
@article{arxiv.2605.02739,
title = {Latent Bridge: Feature Delta Prediction for Efficient Dual-System Vision-Language-Action Model Inference},
author = {Yudong Liu and Yuan Li and Zijia Tang and Yuxi Zheng and Yueqian Lin and Qinsi Wang and Yi Li and Shuangjun Liu and Shuai Zhang and Taotao Jing and Dashan Gao and Ning Bi and Jingwei Sun and Yiran Chen and Hai Li},
journal= {arXiv preprint arXiv:2605.02739},
year = {2026}
}