Folding laundry is a deceptively hard robotics problem — and bimanual garment smoothing is harder still. FCBV-Net (arXiv:2508.05153v2, published April 2026) attacks the category-level generalization gap: instead of training a robot to handle one specific garment instance, it predicts the value of coordinated two-arm actions conditioned on visual features that transfer across an entire garment class. The bottleneck it targets is real — most prior systems either overfit to a single item or generalize perception without being able to plan synergistic dual-arm moves.

Source: arXiv cs.RO (Robotics)