Automated loading and dumping for enhanced efficiency (ALDEE)

Important results from the project

The objective of the project was to enhance the efficiency of the loading and unloading processes in construction projects with the help of machine learning (ML) methods. The project met the objective by developing ML solutions that are trained with scale models to use camera vision for the automation of wheel loaders used for short cycle loading. The project has also developed a simulation environment (aka AI gym) with a model of a Volvo wheel loader where camera-based navigation based on ML can be investigated.

Expected long term effects

Project results include two conference papers on using scale models to train ML models for machine vision, a journal article including a literature review on the importance of ML and computer vision for wheel loader automation, as well as online presentations and a webinar by LTU for Volvo. Additionally, a licentiate thesis on automated loading and unloading solutions based on machine learning has been delivered. With that, the project has contributed with new knowledge in the area, as well as paved the way for continued work in project VALD (financed by FFI).

Approach and implementation

The implementation of the project has been characterized by the Covid-19 situation, which affected the entire project period. Restrictions on traveling and to meet largely made it impossible for LTU to visit and work with full-scale experiments at Volvo CE in Eskilstuna, which was planned for the project. The project has dealt with the situation by focusing on work and approaches that can be used without such experiments, i.e., scale models and simulations. This has led to results completely in line with the project goals and have paved the way for the follow-up project VALD.