FoldBack: Self-Correcting Masked Generative Policy for Long-Horizon Garment Folding
FoldBack retries failed garment grasps before the robot commits to the rest of the fold.
The paper frames the problem as long folding policies continuing after a missed or slipped grasp. FoldBack verifies pick-and-place steps, rolls the robot back to a retryable pre-grasp state, and regenerates the failed segment while avoiding the bad grasp location. On 33 real garments across six categories, it reports 75.2% final folding success, compared with 45.7% for the strongest prior baseline. ArXiv · AI/CL/LG's note
The paper frames the problem as long folding policies continuing after a missed or slipped grasp. FoldBack verifies pick-and-place steps, rolls the robot back to a retryable pre-grasp state, and regenerates the failed segment while avoiding the bad grasp location. On 33 real garments across six categories, it reports 75.2% final folding success, compared with 45.7% for the strongest prior baseline. ArXiv · AI/CL/LG's note
score 5