LunarAssembly

Recent scientific discoveries about the presence of water-ice on moon have led to an increased interest in lunar space missions. For the construction and maintenance of future lunar bases, robotic technologies will become crucial. Teleoperation of robotic systems is an established technology that allows an operator to control a remote robot from a dislocated operator station. The level of emersion of the operator in the remote scene can be significantly increased by the use of haptic feedback. However, for ground-based teleoperation of a lunar robot, significant time delays in the order of several seconds occur, which prevent nowadays teleoperation technology to be applied in such scenarios. While effective control frameworks are available to handle delays up to ~1s, the performance of these algorithms drastically degrades when the delay increases. Motivated by the vision of future lunar bases, our main research challenge is the question how to realize effective teleoperation for large time delays up to the order of 3s. Our approach is based on three scientific pillars: extended reality techniques, machine learning for skill acquisition, and haptic teleoperation. First, we use extended reality techniques in order to obtain an immersive model of the remote environment. Based on these models, we will perform manipulation tasks in a local lab environment, where skills will be learned from demonstrations by teleoperation without time delay. The skills will then be transferred to the remote system and operated in a shared control framework, where the autonomous skills are augmented by direct haptic control from the operator in order to react to unforeseen situations and model errors. During remote control, the control algorithms must preserve stability under large time delays while aiming for high transparency. The LunarAssembly project treats a fundamental scientific challenge, namely the teleoperation of robotic systems under large time delays, motivated by a scenario of remote assembly on the Moon. We believe that such high delays require novel paradigms for teleoperation with an appropriate interplay between autonomous functions and human intervention. We will develop fundamental extensions of the time-domain passivity framework with the aim of improving transparency under high time delay. These novel control algorithms are combined with a curriculum-based framework for incremental learning of autonomous skills, leveraging recent progress in ML for robotics. All this will be enabled by XR technologies based on 3D reconstruction of the remote environment. Besides the progress in these individual research fields, we believe that an innovation also lies in the appropriate interplay of these technologies.