Reviewed by Lexie CornerOct 23 2024
In a new EU project, researchers and industry partners are collaborating to develop reliable and efficient AI-powered machines that minimize risks to both humans and the environment.
The initiative brings together ten European universities and companies, including Umeå University and Umeå-based firms Algoryx and Komatsu Forest, with the aim of combining computational physics methods and cutting-edge AI technology to create safe, autonomous systems that can be practically deployed.
Heavy mobile machinery is widely used in industries such as mining, forestry, agriculture, and construction. However, there is a shortage of operators in many sectors, making it crucial to reduce the environmental footprint of these machines. Increasing automation and improving machine efficiency are, therefore, top priorities.
Must be Predictable
Despite the potential benefits, self-driving heavy machinery presents many complex challenges. These machines are large, powerful, and designed to physically manipulate their environments, meaning they must be not only efficient but also safe and reliable -- what researchers call "predictable. At the same time, they must be able to adapt to unexpected changes in the environment quickly.
The XSCAVE research project aims to address this balance between safety and adaptability. The project's outcomes will be tested on forest machines operating in rugged terrain, earthmoving equipment that unexpectedly encounters large embedded rocks, and outdoor logistics robots facing extreme 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.
Martin Servin, Associate Professor, Umeå University
Informed About Cause and Effect
Instead, the researchers want to provide AI models with more direct information about machine and environmental physics. By integrating mathematical constraints and physics-based models, the AI will learn patterns consistent with fundamental laws of physics, such as energy, inertia, and forces.
Servin concluded, "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."
Source:
Umeå University