Algoryx has participated in a groundbreaking project, in collaboration with Umeå University and Epiroc, to train a deep neural network to control an Epiroc ScoopTram ST18 for autonomous driving and loading, with a minimum of energy consumption.
The project used a full-system simulation of the wheel loader, in an underground mining environment, to train the AI-controller to drive and efficiently empty a pile of muck. The controller was trained to adapt to variable pale shapes, loading the bucket optimally, and still avoiding collisions and wheel-slip.
The physics simulations were made in AGX Dynamics for Unity together with 3D-graphics and sensor simulation in Unity, and ML-Agents for machine learning of the AI-controller.
The project was co-funded by Vinnova SMIG.
The project and results are presented in more detail in the preprint: “Continuous control of an underground loader using deep reinforcement learning” by S. Backman, D. Lindmark, K. Bodin, M. Servin, J. Mörk, and H. Löfgren.
Please don’t hesitate to contact us if you are interested in a similar AI-project, if you want to know more about this project, or want to apply for a free trial license of AGX Dynamcis for Unity.