Closing the Lab-to-Store Gap: A Data-Efficient Post-Training and Experience-Driven Learning VLA Framework for Retail Humanoids

2026-07-22

AI Summary

This work addresses the performance gap experienced by vision-language-action (VLA) humanoid robots when transitioning from controlled laboratory settings to real-world retail environments—a gap caused by execution errors, distribution shifts, and environmental variations. To bridge this divide, the authors propose the DEED framework, which leverages efficient data post-training, experience-based policy optimization, and latent-space behavior analysis to demonstrate that the core issue stems from system integration rather than model architecture. By incorporating control-frequency alignment, task-relevant visual highlighting, textual advantage prompting, and a vision-language value function, DEED transforms a previously failed fine-tuned policy into a high-performance system capable of reliably restocking chips in an actual supermarket. Remarkably, this is achieved using only a single GPU, substantially enhancing real-world deployability.

Abstract

Closing the gap between benchmark performance and reliable real-world operation remains a central challenge for Vision-Language-Action (VLA) humanoid robots, which must handle execution errors, distribution shifts, and environmental variability. This paper presents DEED (Data-Efficient Post-Training and Experience-Driven Learning), a systems-level approach evaluated on a supermarket chip-restocking task using a Unitree G1-Edu humanoid robot and the GR00T N1.6 foundation model. DEED comprises three key components: (1) a data-efficient post-training pipeline with control-frequency alignment, data curation, task-relevant visual highlighting, and reduced VLA dependence; (2) a real-world study of experience-driven refinement, adapted from RECAP via a text-based advantage prefix and a vision-language value function; and (3) a latent-space analysis tool for studying in- and out-of-distribution behavior. Our results suggest that bridging the lab-to-store gap is primarily a systems integration challenge rather than an architectural one: careful data design and targeted post-training can transform a policy that fails under naive fine-tuning into a competent real-world system using only a single GPU.

Problem

• Vision-Language-Action
• lab-to-store gap
• real-world deployment
• distribution shift
• humanoid robots

Innovation

• Vision-Language-Action
• data-efficient post-training
• experience-driven learning
• humanoid robots
• distribution shift