Interactive OAISYS: A photorealistic terrain simulation for robotics research

Abstract

Photorealistic simulation pipelines are crucial for the development of novel robotic methods and modern machine vision approaches. Simulations have been particularly popular for generating labeled synthetic data sets, which otherwise would require vast efforts of manual annotation when using real data. However, these simulators are usually not interactive, and the data generation process cannot be interrupted. Therefore, these simulators are not suitable for evaluating active methods, such as active learning or perception aware path planning, which make decisions based on the observed perception data. In order to address this problem, we propose a modified version of the simulator OAISYS, a photorealistic scene simulator for unstructured outdoor environments. We extended the simulator in order to use it in an interactive way, and implemented a developer-friendly RPC interface so that it is easy for any environment to integrate into the simulator. In this paper, we demonstrate the functionality of the extension on 3D scene reconstruction to show its future research potential and provide an example of the implementation using the middleware ROS. The code is publicly available under https://github.com/DLR-RM/oaisys