Simulator utilization and versatility evaluation of EIPL, a library using deep predictive learning

Abstract

In recent years, the tasks required of robots have become increasingly diverse, and the robots themselves must be able to flexibly adapt to the tasks. Against this background, EIPL has been released as a library for easy use of deep predictive learning. However, some adjustments are required when applying EIPL to robots other than samples. In this paper, we present guidelines for using EIPL, a deep predictive learning library, with various robots, as well as examples of applications. In addition, we also show the evaluation results on a dynamics simulator that was constructed to be used EIPL without a robot.