用轻量模型预测视觉语言动作模型特征变化，大幅减少计算调用
Latent Bridge: Feature Delta Prediction for Efficient Dual-System Vision-Language-Action Model Inference
- 通过预测视觉语言模型输出的增量，避免每步重复计算
- 在多个任务上保持95%-100%性能，减少50%-75%的模型调用次数
- 适配不同架构模型，适合追求推理效率的机器人系统研发
双系统视觉-语言-动作（VLA）模型在机器人操作中表现优异，但受限于视觉语言模型（VLM）主干必须在每个控制步骤执行，产生大量时间冗余特征。本文提出轻量级的Latent Bridge模型，用于预测时间步之间VLM输出的差异，使动作头可基于预测结果运行，而昂贵的VLM主干仅需周期性调用。我们在两种架构迥异的VLA上实现该方法：GR00T-N1.6（特征空间桥接）和π0.5（KV缓存桥接），验证其跨设计泛化能力。采用任务无关的DAgger训练流程，无需修改即可跨基准迁移。在四个LIBERO套件、24个RoboCasa厨房任务及ALOHA模拟立方体转移任务中，Latent Bridge实现95%-100%性能保留，同时减少50%-75%的VLM调用，带来1.65-1.73倍的单次任务净提速。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。