Independent Researcher Achieves Second Place in Real-World Robot Cloth Folding Challenge
Overview
An independent researcher published a technical report on June 2, 2026 (arXiv:2606.27163) detailing a reinforcement learning system that achieved first place in the simulation track and second place in the real-robot track of the LeHome Challenge 2026 cloth folding competition. The system combines behavior cloning with a 'flywheel' training loop involving parallel simulators, reward shaping via intermediate milestones, self-critique heads within a Vision-Language-Action (VLA) model, advantage-weighted sampling strategies, and aggressive visual augmentation to enable sim-to-real transfer. After three months of simulation training followed by one week of real-world fine-tuning on actual hardware, the approach successfully folded garments under strict 'all-or-nothing' geometric criteria.
Generated from attributed reports · Updated 1 hours ago
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Environment: simulation
“<image="https://https://zhida.zhihu.com/share/7703449410511723354/index.html"> 比赛的舞台是NVIDIA的Isaac Lab仿真环境，也就是一个高精度的虚拟物理世界，里面有两只各有六个关节的机械臂，共12个可控维度，每秒做出30个动作决策。”
Exact source · revision 2Source owner not reported
Report timeline
Follow attributed reports and material updates.
- Chinese robotics — Research news discoverySignalIndependent Researcher Achieves Second Place in Real-World Robot Cloth Folding Challenge
An independent researcher published a technical report on June 2, 2026 (arXiv:2606.27163) detailing a reinforcement learning system that achieved first place in the simulation track and second place in the real-robot track of the LeHome Challenge 2026 cloth folding competition. The system combines behavior cloning with a 'flywheel' training loop involving parallel simulators, reward shaping via intermediate milestones, self-critique heads within a Vision-Language-Action (VLA) model, advantage-weighted sampling strategies, and aggressive visual augmentation to enable sim-to-real transfer. After three months of simulation training followed by one week of real-world fine-tuning on actual hardware, the approach successfully folded garments under strict 'all-or-nothing' geometric criteria.
Event coverage history
There is not enough continuous observation data to show a trend.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).