TR2026-174

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

    •  Ma, C., Yang, Y., Corcodel, R., Jain, S., Wu, A., Hori, C., Romeres, D., "FurnitureVLA: Learning Long-Horizon Bimanual Furniture Assembly with Vision-Language-Action Model", Conference on Robot Learning (CoRL) Workshop on The Action Gap: Large Generative Models Learn Rich Knowledge, November 2026.
      BibTeX TR2026-174 PDF
      • @inproceedings{Ma2026nov,
      • author = {Ma, Chenyang and Yang, Yue and Corcodel, Radu and Jain, Siddarth and Wu, Andrew and Hori, Chiori and Romeres, Diego},
      • title = {{FurnitureVLA: Learning Long-Horizon Bimanual Furniture Assembly with Vision-Language-Action Model}},
      • booktitle = {Conference on Robot Learning (CoRL) Workshop on The Action Gap: Large Generative Models Learn Rich Knowledge},
      • year = 2026,
      • month = nov,
      • url = {https://www.merl.com/publications/TR2026-174}
      • }
  • MERL Contacts:
  • Research Areas:

    Artificial Intelligence, Machine Learning, Robotics, Speech & Audio

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 that jointly predicts actions and a continuous progress signal, enabling automatic subtask transitions and reducing compounding errors. FurnitureVLA improves average simulation success from 48% to 80% over baselines across three furniture types, with a further 21% gain from our study of perception and control design factors, and transfers to a real Kinova Gen3 platform with only 16% drop on the hardest task.