Latent Bridge: Feature Delta Prediction for Efficient Dual-System Vision-Language-Action Model Inference

TL;DR

Latent Bridge predicts feature deltas in vision-language-action models to reduce expensive backbone calls, maintaining high performance while significantly speeding up inference.

Contribution

Introduces Latent Bridge, a lightweight delta prediction model that enables efficient VLA inference across various architectures and benchmarks.

Findings

Achieves 95-100% performance retention with 50-75% fewer VLM calls.

Yields 1.65-1.73x net per-episode speedup.

Demonstrates generalization across multiple VLA architectures and tasks.

Abstract

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.