Image: Canva professional
Research project The Artemis project involves establishing a permanent base on the Moon. It is to be constructed using lunar soil, regolith, and construction machines. How these machines should be designed and controlled is still largely unknown and will be subject to intensive research in digital environments. This project aims to develop key simulation and AI solutions for the development platform.
The project develops technology for physics-based digital twins of mobile and autonomous robotic systems on the Moon based on available data on the lunar surface and soil properties. By realistic simulation of the interaction between equipment and regolith in lunar terrain, we address previously unsolved problems in mechanical design, autonomous navigation and control.
Numerous projects worldwide aim to explore the Moon, investigate the potential for lunar mineral exploitation, and establish bases on the lunar surface. The success of these endeavours hinges on vehicles capable of navigating and traversing the largely unknown lunar terrain. These vehicles must also perform complex robotic tasks, such as digging, drilling, moving regolith, and constructing and assembling structures. These vehicles must be extremely reliable with a high degree of autonomy.
Engineering these machines presents a significant challenge since fully realistic lunar lab environments cannot be replicated on Earth. Physics-based full system simulation is a crucial method that enables iterative design, training of physics-informed AI controllers, and the ability to validate and verify engineered solutions across millions of different scenarios and circumstances.
The goal of the project is to develop a state-of-the-art framework for physics-based digital twins of mobile lunar robotic systems as well as essential components in the autonomy stack, based on all existing data for the Moon.
The software will realistically visualize the Lunar surface using Lunar Reconnaissance Orbiter (LRO) data. Dynamic regolith-vehicle interaction will be simulated using hybrid particle-mesh discretization, constitutive physics models, a semi-implicit variational time stepper, and high-performance GPU accelerated solvers.
An autonomy stack will be developed that support teleoperation, AI-assisted operation, to full AI-based autonomy for navigation and control. Physical AI foundational models and differential physics-based controllers will be developed in parallel initiatives, but the autonomy stack will incorporate such interfaces. The project will integrate and adapt our state-of-the-art research on physical AI for navigation and control in lunar conditions.
