Reproduce “FLIP: Flowability-Informed Powder Weighing” 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

Nikola Radulov, Alex Wright, Thomas Little, Andrew I. Cooper, Gabriella Pizzuto

FLIP (Flowability-Informed Powder Weighing) integrates real-world powder flowability data into simulation parameter optimization and curriculum learning to enhance robotic powder weighing. The framework enables robots to achieve an average error of 2.12 ± 1.53 mg in real-world scenarios, demonstrating improved accuracy and generalization across diverse powders and untrained target weights.

Key insights

  • Integrating real-world flowability data (AoR) to calibrate simulation micro-parameters significantly improves the fidelity of simulated granular materials, accurately reproducing macroscopic powder behavior.
  • Explicitly leveraging material properties, such as flowability, within a curriculum learning strategy enhances the efficiency and robustness of robotic policy acquisition for complex material handling tasks.
  • An 'easy-to-hard' training progression based on material flowability leads to stronger generalization capabilities for robotic manipulation, extending to previously unseen materials and target weights.

Summary by alphaXiv.

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