FurnitureVLA: Learning Long-Horizon Bimanual Furniture Assembly with Vision-Language-Action Model

2026-07-01

AI Summary

This work addresses the challenge of long-horizon furniture assembly at real-world scale, a task that existing dual-arm robotic approaches struggle with due to their typical confinement to toy-scale setups or single-arm manipulation. The paper presents the first real-scale bimanual robotic assembly system based on a vision-language-action (VLA) model, introducing a novel progress-augmented VLA architecture. High-quality demonstration data are collected via VR teleoperation, and the system jointly predicts semantic subtask labels and continuous progress signals to enable automatic subtask switching and error suppression. Evaluated on three furniture assembly tasks, the approach improves simulation success rates from 48% to 80%, with an additional 21% gain achieved through optimized perception and control design. Real-world experiments on a Kinova Gen3 platform demonstrate robust performance, with the most challenging task showing only a 16% drop in success rate—significantly outperforming current methods.

Abstract

Current work on robot furniture assembly mostly focuses on toy-scale settings or single-arm manipulation. We introduce FurnitureVLA, the first systematic study of real-scale bimanual furniture assembly using Vision-Language-Action models (VLAs). We formalize the task, develop a scalable simulation pipeline for expert data generation and evaluation, and build a VR teleoperation system for single-operator bimanual control to collect high-quality real-world demonstrations. To address extreme long-horizon assembly with up to 7 subtasks and 1550 control steps, we propose a progress-enhanced VLA, finetuned on semantically grounded subtasks, that jointly predicts actions and a continuous progress signal, enabling automatic subtask transitions and reducing compounding errors during inference. We further study perception and control design factors that critically affect precision in real-scale assembly. FurnitureVLA improves average simulation success from 48% to 80% compared to baselines across three furniture types, with an additional 21% gain from our design factor study. We validate on a real Kinova Gen3 platform with only 16% drop on the hardest task.

Problem

• bimanual manipulation
• furniture assembly
• long-horizon task
• real-scale robotics
• Vision-Language-Action models

Innovation

• Vision-Language-Action Model
• Bimanual Manipulation
• Long-Horizon Task
• Progress Signal
• VR Teleoperation