Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note2 min read

Learning to Fold: prizewinning solution at LeHome Challenge 2026 (1st place online, 2nd offline)

Source

arXiv

arxiv.org/abs/2606.27163v1

Read the source

Source

Plain-Language Brief

The text describes a prize-winning system for bimanual garment folding from the LeHome Challenge 2026. The system used a reinforcement learning loop to improve a vision-language-action (VLA) policy, placing first in the online round and second in the real-world final. The VLA policy predicts actions and success, progress, and future quantities that drive advantage estimation, failure detection, and candidate selection.

What the Document Says

  • The text is about a solution to the LeHome Challenge 2026, an ICRA 2026 competition on bimanual garment folding.
  • The system placed 1st in the online (simulation) round and 2nd in the real-world final.
  • It improves a vision-language-action (VLA) policy using a reinforcement-learning loop.
  • The VLA policy predicts actions, success, progress, and a few task-relevant future quantities.
  • These predictions drive advantage estimation, live failure detection, and candidate selection.

Key Concepts Mentioned

  • LeHome Challenge 2026
  • ICRA 2026
  • bimanual garment folding
  • vision-language-action (VLA) policy
  • reinforcement learning
  • simulation
  • real-world final
  • advantage estimation
  • failure detection
  • candidate selection

Claims Made by the Document

  • The described system won the LeHome Challenge 2026.
  • The system improved a VLA policy using reinforcement learning.

Evidence Used

  • The text mentions that the system placed first in the online round and second in the real-world final.
  • The text explains that the system improves a VLA policy with a reinforcement-learning loop.

Methods / Approach

  • Use of a vision-language-action (VLA) policy with reinforcement learning
  • Asynchronous distributed training / rollout pipeline
  • Inference-time hyperparameters optimization via Thompson sampling
  • Sim-to-real recipe with camera-alignment tooling, heavy augmentation and DAgger-like HIL data collection

Limitations Stated by the Authors

The text does not mention any limitations by the authors.

Limitations Noted by the Agent

Agent interpretation — not from the source.

The system might have limitations in handling complex or unfamiliar garments due to its focus on bimanual garment folding. Additionally, it may encounter challenges when adapting to different lighting conditions or garment materials.

Open Questions

  • How does the system handle different garment types?
  • How does the system adapt to changes in lighting conditions?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The text only discusses a specific application of the system (bimanual garment folding), so it is unclear how well it would perform in other tasks or contexts.

Raw Abstract / Excerpt

I describe my solution to the LeHome Challenge 2026, an ICRA 2026 competition on bimanual garment folding. The system placed 1st of 62 teams in the online (simulation) round and 2nd in the real-world final. It improves a vision-language-action (VLA) policy with a reinforcement-learning loop. The policy is its own value function: the same network that predicts actions also predicts success, progress, and a few task-relevant future quantities, and those predictions drive advantage estimation, live failure detection, and candidate selection. The work mostly recombines existing RL ideas with engineering and optimization contributions that can be used together as one recipe or individually: AWR + RECAP combined for flow-matching VLA; an asynchronous distributed training / rollout pipeline through HuggingFace Hub; inference-time hyperparameters optimization via Thompson sampling; a sim-to-real recipe with camera-alignment tooling, heavy augmentation and DAgger-like HIL data collection.