RA-VLA: Retrieval-Augmented VLA for Test-Time Adaptation

Vision-Language-Action (VLA) models provide a versatile foundation for general robotic manipulation, yet they exhibit significant brittleness when confronted with novel task distributions. While In-Context Imitation Learning (ICIL) offers a training-free alternative, existing frameworks suffer from an adaptation bottleneck that hinders the effective translation of expert context to executable actions. This failure originates from superficial retrieval mechanisms and an inherent behavioral inertia that anchors the policy to its pre-trained priors. To address these limitations, we present RA-VLA, a retrieval-augmented VLA framework that integrates behavior-aligned context retrieval with a grounded execution pipeline. By enforcing faithful adherence to functional cues within a scalable architecture, RA-VLA facilitates seamless task adaptation while preserving inference efficiency. Our empirical evaluations across the LIBERO benchmark and a real-world UR5e environment demonstrate that RA-VLA achieves superior success rates and computational efficiency, establishing a robust framework for training-free robotic adaptation.

While modern robots handle everyday instructions well, they often freeze when encountering unfamiliar tasks. Even if we show them quick examples to help them adapt, they tend to blindly fall back on old habits instead of learning from what they were just shown. To fix this, we developed RA-VLA, a smart memory assistant. It helps the robot recall the best examples from its library and faithfully follow those actions without sacrificing its processing speed. Across both virtual and real-world environments, RA-VLA allows the robot to complete brand-new tasks far more reliably and quickly. This brings us closer to versatile robots that can instantly master new chores in homes and workspaces just by observing us.