Towards Coordinated Multirobot Missions for Lunar Sample Collection in an Unknown Environment

Abstract

In this paper, the recent results of the space project IMPERA are presented. The goal of IMPERA is the development of a multirobot planning and plan execution architecture with a focus on a lunar sample collection scenario in an unknown environment. We describe the implementation and verification of different modules that are integrated into a distributed system architecture. The modules include a mission planning approach for a multirobot system and modules for task and skill execution within a lunar use‐case scenario. The skills needed for the test scenario include cooperative exploration and mapping strategies for an unknown environment, the localization and classification of sample containers using a novel approach of semantic perception, and the skill of transporting sample containers to a collection point using a mobile manipulation robot. Additionally, we present our approach of a reliable communication framework that can deal with communication loss during the mission. Several modules are tested within several experiments in the domain of planning and plan execution, communication, coordinated exploration, perception, and object transportation. An overall system integration is tested on a mission scenario experiment using three robots.

Review of Multi-Agent Algorithms for Collective Behavior: a Structural Taxonomy

“…Multi-agent reinforcement learning (MARL) has been used for exploration (Chalkiadakis and Boutilier, 2003) and task allocation (Liu and Nejat, 2016). Frontier techniques (Burgard et al, 2000) are used for urban search-and-rescue, reconnaissance (Olson et al, 2012) and sample collection (Eich et al, 2014). Network flow formulations have been proposed for Air Traffic Control (Menon et al, 2004) and for control of autonomous vehicles offering on-demand transportation (Pavone et al, 2012;Rossi et al, 2018).…”

Towards Autonomous Planetary Exploration

“…The experiences gathered with the successful and ongoing Mars rover missions like the Mars Exploration Rover Mission (MER) [46] and the Mars Science Laboratory (MSL) [28] as well as earlier considerations on rover autonomy [69] clarify requirements and space-suitable options regarding hardware as well as software components. Autonomous navigation solutions for unstructured and unknown environments taking robot safety, resource management (e. g., power consumption) and general robustness into account are available in many robotic research areas such as autonomous driving, e. g., [67,76], search and rescue [57] and planetary rovers / field robotic systems tested on Earth [26,43,60,65,73]. These systems, including all systems participating in the SpaceBotCamp 2015 challenge, use a variety of optical sensors for navigation, most commonly laser scanners and active RGB-D cameras [14,66] or a combination of those, e. g., [32,60,65].…”

“…The advantages of camera systems are the availability of both, mature algorithms and compact, low-power flight-qualified cameras, whereas flight-qualified versions of other sensors for navigation and mapping, like suitable laser scanners, are currently not available [28,46]. Sensor-specific noise characteristics and a typically less dense depth reconstruction of passive stereo camera systems pose a higher challenge on navigation and mapping algorithms compared to laser scanner-based systems [26,60,65], which allow highprecision measurements within a longer range of distances. Similar to past and current Mars exploration missions, the rover described in [73] as well as our own system only employ stereo camera systems.…”

Robot Teams for Intervention in Confined and Structured Environments

“…Moreover, some important aspects such as communications problems and real‐time requirements are not taken into account. Some frameworks have been proposed for working in other environments, such as the surface of the Moon (Eich, Hartanto, Kasperski, Natarajan, & Wollenberg, ), but these also lack real‐time aspects and stable communication links needed for a distributed control loop.…”