AsyncShield: A Plug-and-Play Edge Adapter for Asynchronous Cloud-based VLA Navigation

TL;DR

AsyncShield is a plug-and-play framework that enhances cloud-based VLA robot navigation by converting temporal delays into spatial corrections, improving safety and success rates without model fine-tuning.

Contribution

It introduces a deterministic white-box spatial mapping and reinforcement learning-based adaptation to address latency issues in cloud-based VLA navigation.

Findings

Improves navigation success rate in simulation and real-world tests.

Enhances physical safety by effectively avoiding collisions.

Operates as a lightweight, plug-and-play module without model fine-tuning.

Abstract

While Vision-Language-Action (VLA) models have been demonstrated possessing strong zero-shot generalization for robot control, their massive parameter sizes typically necessitate cloud-based deployment. However, cloud deployment introduces network jitter and inference latency, which can induce severe spatiotemporal misalignment in mobile navigation under continuous displacement, so that the stale intents expressed in past ego frames may become spatially incorrect in the current frame and lead to collisions. To address this issue, we propose AsyncShield, a plug-and-play asynchronous control framework. AsyncShield discards traditional black-box time-series prediction in favor of a deterministic physical white-box spatial mapping. By maintaining a temporal pose buffer and utilizing kinematic transformations, the system accurately converts temporal lag into spatial pose offsets to restore the VLA's original geometric intent. To balance intent restoration fidelity and physical safety, the edge adaptation is formulated as a constrained Markov decision process (CMDP). Solved via the PPO-Lagrangian algorithm, a reinforcement learning adapter dynamically trades off between tracking the VLA intent and responding to high-frequency LiDAR obstacle avoidance hard constraints. Furthermore, benefiting from a standardized universal sub-goal interface, domain randomization, and perception-level adaptation via Collision Radius Inflation, AsyncShield operates as a lightweight, plug-and-play module. Simulation and real-world experiments demonstrate that, without fine-tuning any cloud-based foundation models, the framework exhibits zero-shot and robust generalization capabilities, effectively improving the success rate and physical safety of asynchronous navigation.