Requirement Analysis for Perception on Mobile Assistant Robot Platforms in Multi-Modal Environment Conditions

Kurzfassung

Human spaceflight beyond low earth orbit has gained more and more interest in recent times. Long-running missions and exploration of extra-terrestrial bodies will need the assistance of robotic systems during operation. In order to reduce the workload for astronauts, robots have to be equipped with autonomous features. The main requirement for this is a robust and accurate working localization and perception system. Especially in collaboration with on-site human operators and other participating robots, establishing situation awareness is essential in operation safety. At the German Aerospace Center (DLR), Surface Avatar and SMiLE2gether are two projects in the field of humanoid service and assistance robots, with a particular focus on autonomous robot operation. In this work, the focus is directed to examining mission-related environments with the consideration of available external circumstances for the realization of a robust perception framework. Starting from the different branches of the human perception system, their viability in the machine world is evaluated, and promising approaches are being transferred into the field of robotics. The available robotic hardware sets additional boundary conditions. Therefore, we focus on the visual and auditory domains for detailed examination. Visual environment conditions are analyzed for Visual Odometry, Mapping, and Localization modules of navigation systems. Distinct characteristics are identified and evaluated to enhance the system's robustness against environmental influences and perturbations. Auditory conditions are considered to be used for other perception modules and the enhancement of the navigation pipeline. In particular, beamforming for sound source localization and sound classification will be analyzed for event detection systems. For this purpose, we extract fundamental properties of the given mission-related environments and describe the appearance of the scenes. Particular interest will be given to the spatial information gained across the scene for the presence and absence of specific properties. Furthermore, current state-of-the-art visual approaches are evaluated in an experimental setup for their usability and use on mobile platforms with limited computational resources. Therefore, datasets from the mission-related surroundings are recorded to provide an adequate basis for the evaluation. Finally, findings and realizations from each examination process are summarized in recommendations for improving multi-modal perception systems within the considered projects.