Closing the Lab-to-Store Gap: A Data-Efficient Post-Training and Experience-Driven Learning VLA Framework for Retail Humanoids
DEED turns a naively failing humanoid restocking policy into a working supermarket task system with targeted post-training on one GPU.
The paper tests a Unitree G1-Edu humanoid on chip restocking using GR00T N1.6. Its framework combines curated data, control-frequency alignment, visual highlighting, reduced dependence on the VLA model, and experience-driven refinement. The authors argue the remaining gap is less about new robot architectures than careful systems integration for real-world variability. ArXiv · AI/CL/LG's note
The paper tests a Unitree G1-Edu humanoid on chip restocking using GR00T N1.6. Its framework combines curated data, control-frequency alignment, visual highlighting, reduced dependence on the VLA model, and experience-driven refinement. The authors argue the remaining gap is less about new robot architectures than careful systems integration for real-world variability. ArXiv · AI/CL/LG's note
score 4