2019/04/04 by Tatiana López-Guevara, Lopez-Guevara, Tatiana, Rita Pucci +10
Computer Science · Physics and Astronomy · Engineering · #Reinforcement Learning in Robotics #Micro and Nano Robotics #Modular Robots and Swarm Intelligence
paper · pdf · doi:10.48550/arxiv.1904.02431
Our brains are able to exploit coarse physical models of fluids to solve\neveryday manipulation tasks. There has been considerable interest in developing\nsuch a capability in robots so that they can autonomously manipulate fluids\nadapting to different conditions. In this paper, we investigate the problem of\nadaptation to liquids with different characteristics. We develop a simple\ncalibration task (stirring with a stick) that enables rapid inference of the\nparameters of the liquid from RBG data. We perform the inference in the space\nof simulation parameters rather than on physically accurate parameters. This\nfacilitates prediction and optimization tasks since the inferred parameters may\nbe fed directly to the simulator. We demonstrate that our "stirring" learner\nperforms better than when the robot is calibrated with pouring actions. We show\nthat our method is able to infer properties of three different liquids --\nwater, glycerin and gel -- and present experimental results by executing\nstirring and pouring actions on a UR10. We believe that decoupling of the\ntraining actions from the goal task is an important step towards simple,\nautonomous learning of the behavior of different fluids in unstructured\nenvironments.\n