NEWS When heavy machinery is to become robots with autonomous capabilities, safety is at highest priority. In a new EU project, researchers and industry collaborate to develop reliable and efficient AI-driven machines that minimize the risk of harming people or the environment.
Researchers at Umeå University are developing autonomous systems that combine the laws of physics with advanced AI.
ImageMartin ServinTen European universities and companies, including Umeå University and the Umeå companies Algoryx and Komatsu Forest, are participating in the project where computational science meets the latest AI technology. The goal is to combine methods in computational physics with artificial intelligence, AI, to create safe and efficient autonomous systems that work in practice.
Heavy mobile machinery plays an important role in many industries, such as mining, forestry, agriculture, and construction, but in many places, there is a shortage of operators, and the need to reduce the machines' environmental footprint is urgent. Increased automation and efficiency of the machines are therefore high on the agenda.
Self-driving machines, however, are associated with several difficult challenges. They are heavy and powerful – designed to physically manipulate their surroundings. Therefore, the systems must be safe and reliable, i.e., predictable. At the same time, they must have the ability to quickly adapt to sudden changes in the environment.
Self-driving machines need to be able to adapt to changes in the environment in order not to risk harming the environment or, in the worst case, people.
ImageMartin ServinBalancing these seemingly contradictory requirements is the goal of the research project XSCAVE. The project's results will be demonstrated on forest machines operating in rough terrain, earthmoving equipment suddenly encountering soil with large embedded rocks, and outdoor logistics robots in challenging weather conditions.
“Today, physics-based simulation is used to test and train control systems and advanced AI models, so-called deep neural networks. The use of simulation is a safe and efficient way to cover a wide range of scenarios, but it remains difficult to ensure a safe behavior in situations that differ significantly from the training cases,” says Martin Servin, who leads Umeå University's research in the project.
Instead, the researchers want to give AI models more direct insight about the physics of the machines and of the environment. By incorporating mathematical constraints and models, they will be able to learn only patterns that are consistent with the fundamental laws of physics regarding energy, inertia, and forces.
“When we embed computational models and equation solvers for the physics, we make the AI informed about cause and effect, and a tool for predicting the probable outcome of planned movements before they are executed. This makes it possible to rule out options associated with an unacceptable risk of damage or negative environmental impact. At the same time, we believe this is a way to achieve higher precision and efficiency,” says Martin Servin.
XSCAVE is acronym for “Explainable, Safe, Contact-Aware Planning and Control for Heavy Machinery Manipulation and Navigation.” The project runs 2025–2028 and is financed with 8 million euro from the EU program Horizon Europe. Participating organizations include Aalto University, Algoryx, Forschungszentrum Informatik, Clevon, Czech Technical University in Prague, Komatsu Forest, Novatron, Tampere University, Toshiba Europe, Umeå University, and University of Tartu.
The Nobel Prize in Physics 2024 is awarded for discoveries and inventions that enable machine learning with artificial neural networks. The laureates John Hopfield and Geoffry Hinton developed methods that give networks the ability to store memories and find underlying patterns in data.
Today, artificial neural networks play a central role in the AI field and are indispensable for, for example, image recognition and generative language models. The XSCAVE project explores new ways to combine physics and artificial neural networks to predict the movement of machines in terrain.
