In a new doctoral thesis from Umeå University, Dr. Koji Aoshima addresses the challenge of achieving autonomous and high-performance earthmoving across a sequence of tasks. The research has been conducted in collaboration with Komatsu Ltd, Japan, and AGX Dynamics from Algoryx was used throughout the project to generate high-fidelity synthetic data for deep learning.
On 24 January 2025, Dr. Koji Aoshima successfully defended his PhD thesis, “High-Performance Autonomous Wheel Loading: A Computational Approach,” at Umeå University, Sweden.
Dr. Aoshima has conducted his doctoral research at Umeå University’s UMIT Research Lab, in collaboration with Komatsu Ltd., Japan, under the supervision of Associate Professor Martin Servin, Professor Eddie Wadbro, and Takehiro Komatsu.
Highlighting the critical role of physics simulation
In his thesis, Dr. Aoshima addresses the challenge of achieving autonomous and high-performance earthmoving across a sequence of tasks. The core research questions addressed in the thesis include:
- How can real-time simulation support the development of autonomous wheel loaders?
- How should world models be structured to account for evolving pile states and sequential actions?
- What optimization framework enables adaptive and high-performance loading?
- How can look-ahead tree search methods improve decision-making in dynamic environments?
To answer these questions, Dr. Aoshima developed:
- A real-time wheel loader simulator for testing and validation
- World models that incorporate sequential loading actions and pile dynamics
- An end-to-end optimization framework for task execution
- A look-ahead tree search method for efficient action selection
Throughout this PhD project, Dr. Aoshima and his research partners have leveraged AGX Dynamics from Algoryx, utilizing non-smooth contacting multibody dynamics to model both the vehicle and the soil.
Dr Koji Aoshima’s research makes a significant contribution to the field of autonomous earthmoving while demonstrating the critical role of high-fidelity physics simulation in achieving this goal.
Examined by academic experts in AI and robotics
The public defense of the thesis was held at Umeå University, Sweden.
Professor Reza Ghabcheloo from Tampere University, Finland, served as the external examiner for the public defense.
The examining committee consisted of:
- Markus Bohlin, Mälardalens University, Professor of computer science, AI and product realization.
- Shiva Sander Tavallaey, adj professor KTH, senior principal scientist at ABB Corporate Research in Västerås and AI-lead in Sweden
- Johannes A. Stork, Örebro University, associate professor of computer science specialized in adaptive and interpretable learning systems
References
Koji Aoshima: “High-performance autonomous wheel loading: a computational approach” (Doctorat thesis, Umeå University, 2025)