Development of a generative model for a mobile robot operated in everyday environments and its evaluation by pilot experiment with data collection

抄録

In recent years, advancements in navigation technology have facilitated the realization of robots capable of robust path planning in structured spaces. However, especially in human coexisting environments, a framework for robust mobility needs to be developed to cope with unpredictable dynamic changes, such as the movements of surrounding individuals. Therefore, we constructed a generative model for a series of observations and actions collected during teleoperated robot control, with the aim of transferring recorded behaviors to the robot. In this report, we elaborate on the data collection of first-person perspective images and action sequences using a teleoperation system. In order to examine the practicality, we implemented a naive generative model and investigated the possibility of generating action series and future observations.