Abstract: Co-Evolution of Multi-Robot Controllers and Task Cues for Off-World Open Pit Mining
The Moon has been found to hold vast quantities of critical resources needed to kickstart a future space economy. These resources include water ice, oxygen bound in minerals, and metals that can support fuel production, construction, manufacturing, and sustained lunar operations. The challenge is how to practically extract these resources at large scale under low gravity, abrasive dust, large temperature variations, radiation, and with little or no onsite human support. Robots are ideally suited for these dull, dirty, and dangerous tasks, but the challenge is to scale-up productivity with an ever-increasing number of robots. Our earlier work shows that fleets of autonomous decentralized robots have an optimal operating density. Too few robots result in insufficient labor, while too many robots result in antagonism, where robots interfere with or undo each other’s work and can become stuck in gridlock. In this paper, we explore methods to improve the scalability of multi-robot excavation and resource gathering using Artificial Neural Tissue (ANT), a bio-inspired evolvable neural-network architecture. ANT starts with a blank slate and does not require human-authored operation scripts or detailed modeling of the robot kinematics and dynamics. Instead, controllers evolve through trial and error using a global fitness function, sensory inputs, and a generic set of behavior primitives. We examine how robot density, behavior primitives, templates, stigmergy, and environmental cues affect system performance. The results show that ANT can evolve creative cooperative behaviors, including bucket brigades, that improve material transport and reduce the negative effects of antagonism at higher robot densities. ANT controllers can also exploit simple environmental cues, such as light beacons, particularly under severe time constraints. The work shows that scalable off-world mining can emerge from simple robots, decentralized control, and carefully selected environmental cues, enabling robust, extensible, and increasingly autonomous lunar resource gathering and site preparation.
Bio: Jekan Thanga has a background in aerospace engineering from the University of Toronto and is an Associate Fellow of the AIAA. He worked on Canadarm, Canadarm2, and the DARPA Orbital Express missions at MDA Space Missions. Jekan obtained his Ph.D. in space robotics at the University of Toronto Institute for Aerospace Studies (UTIAS) and completed his postdoctoral training at MIT's Field and Space Robotics Laboratory (FSRL). Jekan Thanga is an Associate Professor and heads the Space and Terrestrial Robotic Exploration (SpaceTREx) Laboratory and the NASA-funded ASTEROIDS (Asteroid Science, Technology and Exploration Research Organized by Inclusive eDucation) Center at the University of Arizona. His research focuses on autonomous robotic systems for space exploration, including lunar and planetary robotics, in-situ resource utilization, autonomous construction of space infrastructure, spacecraft and rover autonomy, distributed space systems, and resilient robotic operations in extreme environments on the Moon, other planetary bodies, and Earth. Jekan and his team of students have co-authored more than 250 technical publications. He and his team of students were winners of the Popular Mechanics Breakthrough Award in 2016 for proposing the SunCube FemtoSat and received the Best Paper Presentation Award at AMOS 2019 for the Early Warning Constellation to Detect Incoming Meteor Threats. Jekan and his team of students were finalists in the NASA 2020 BIG Idea Challenge and winners of the 2021 NASA RASC-AL Competition. In recognition of his contributions to astronautics, the International Astronomical Union named asteroid (20460) Jekanthanga in his honor.