Robot motion planning for pouring liquids

Abstract

We present a new algorithm to compute a collision-free trajectory for a robot manipulator to pour liquid from one container to the other. Our formulation uses a physical fluid model to simulate its highly deformable motion. We present a simulation guided and optimization based method to automatically compute the transferring trajectory. We use the full-featured and accurate Navier-Stokes model that provides the fine-grained information of velocity distribution inside the liquid body. Moreover, this information is used as an additional guiding energy term for the planner. Our approach includes a tight integration between the fine-grained fluid simulator, liquid transfer controller, and the optimization-based planner. We have implemented the method using a hybrid particle-mesh fluid simulator (FLIP) and demonstrated its performance on 4 benchmarks with different cup shapes and viscosity coefficients.