Manipulation of Granular Materials by Learning Particle Interactions

Manipulation of granular materials such as sand or rice remains an unsolved\nproblem due to challenges such as the difficulty of defining their\nconfiguration or modeling the materials and their particles interactions.\nCurrent approaches tend to simplify the material dynamics and omit the\ninteractions between the particles. In this paper, we propose to use a\ngraph-based representation to model the interaction dynamics of the material\nand rigid bodies manipulating it. This allows the planning of manipulation\ntrajectories to reach a desired configuration of the material. We use a graph\nneural network (GNN) to model the particle interactions via message-passing. To\nplan manipulation trajectories, we propose to minimise the Wasserstein distance\nbetween a predicted distribution of granular particles and their desired\nconfiguration. We demonstrate that the proposed method is able to pour granular\nmaterials into the desired configuration both in simulated and real scenarios.\n