K2

K² · Artificial intelligence

Two robot arms build real furniture by tracking their own progress

The authors introduce FurnitureVLA, described as the first systematic study of real-scale bimanual furniture assembly using Vision-Language-Action models.

CM
YY

Chenyang Ma, Yue Yang, Radu Corcodel et al.

7 authors · cs.RO, cs.AI

arXiv preprintArtificial intelligenceJul 2026 · ~70s read

The 30-second scan

Building a bookshelf takes up to 1,550 tiny robot movements strung together across 7 separate steps.

  1. They 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 real-world demonstrations.
  2. The proposed method is a progress-enhanced VLA, finetuned on semantically grounded subtasks, that jointly predicts actions and a continuous progress signal to enable automatic subtask transitions and reduce compounding errors during inference.
  3. The task addressed involves extreme long-horizon assembly with up to 7 subtasks and 1550 control steps.
Up to 7 subtasks and 1550 control steps in the long-horizon assembly taskImproves average simulation success from 48% to 80% compared to baselines across three furniture typesAdditional 21% gain from the design factor study