Robotic Paper Wrapping by Learning Force Control
Hiroki Hanai, Takuya Kiyokawa, Weiwei Wan, Kensuke Harada
TL;DR
This work tackles robotic wrapping of boxes with wrapping paper, a task complicated by material deformation and tear/wrinkle risks. It introduces a hybrid imitation learning and reinforcement learning framework that learns nominal TCP trajectories from human demos and uses reinforcement learning to optimize force-control parameters, guided by a phase-estimation network for per-phase policy selection. Results show substantial reductions in tears and wrinkles and robust generalization across paper thickness and box sizes, validating both trajectory learning and material-adaptive force control. The approach promises a practical, material-agnostic wrapping solution and points to future enhancements with vision and dual-arm coordination.
Abstract
Robotic packaging using wrapping paper poses significant challenges due to the material's complex deformation properties. The packaging process itself involves multiple steps, primarily categorized as folding the paper or creating creases. Small deviations in the robot's arm trajectory or force vector can lead to tearing or wrinkling of the paper, exacerbated by the variability in material properties. This study introduces a novel framework that combines imitation learning and reinforcement learning to enable a robot to perform each step of the packaging process efficiently. The framework allows the robot to follow approximate trajectories of the tool-center point (TCP) based on human demonstrations while optimizing force control parameters to prevent tearing or wrinkling, even with variable wrapping paper materials. The proposed method was validated through ablation studies, which demonstrated successful task completion with a significant reduction in tear and wrinkle rates. Furthermore, the force control strategy proved to be adaptable across different wrapping paper materials and robust against variations in the size of the target object.