Reproduce “ShakingBot: Dynamic Manipulation for Bagging” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Ningquan Gu, Zhizhong Zhang, Ruhan He, Lianqing Yu

Researchers at Wuhan Textile University developed ShakingBot, a robotic framework that utilizes dynamic manipulation to efficiently open crumpled plastic bags, insert items, and lift them for transport. The system achieves improved success rates and efficiency compared to quasi-static methods, demonstrating 6/8 full success in medium difficulty tasks with fewer actions and less time than the AutoBag baseline, and generalizes across bags of varying sizes, patterns, and colors.

Key insights

  • Dynamic manipulation strategies, inspired by human behavior, can dramatically reduce the complexity associated with modeling and reacting to intricate, crumpled configurations of deformable objects like plastic bags, leading to more efficient operations.
  • A perception module trained on depth images using a fully convolutional network architecture provides robustness and generalization capabilities across variations in bag size, pattern, and color, eliminating the need for retraining on new visual features.
  • Specific dynamic action primitives, such as 'Bag Adjustment' for separating bag layers and 'One-arm Holding' for stabilizing the bag during insertion, are critical for enhancing the success rate and robustness of complex deformable object manipulation tasks.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.