Granular Material
Granular materials, encompassing diverse substances like sand, rice, and soil, present unique challenges in modeling their complex behavior and manipulating them robotically. Current research focuses on developing accurate yet computationally efficient simulation methods, often employing graph neural networks, differentiable physics simulators, and optimal transport algorithms to predict and control granular flow. These advancements are crucial for improving robotic manipulation in various fields, including construction, agriculture, and space exploration, as well as for designing novel computing devices leveraging the inherent dynamics of granular metamaterials. The development of large-scale datasets and open-source tools is also driving progress in this area.
Papers
PSDNet: Determination of Particle Size Distributions Using Synthetic Soil Images and Convolutional Neural Networks
Javad Manashti, Pouyan Pirnia, Alireza Manashty, Sahar Ujan, Matthew Toews, François Duhaime
Comparing PSDNet, pretrained networks, and traditional feature extraction for predicting the particle size distribution of granular materials from photographs
Javad Manashti, François Duhaime, Matthew F. Toews, Pouyan Pirnia, Jn Kinsonn Telcy
Cross-Tool and Cross-Behavior Perceptual Knowledge Transfer for Grounded Object Recognition
Gyan Tatiya, Jonathan Francis, Jivko Sinapov