DeepHaul: A Deep Learning and Reinforcement Learning-Based Smart Automation Framework for Dump Trucks


In surface mining operations, the haul truck contributes the largest share to the overall operating cost, along with being a significant contributor to the injuries and fatalities to the operators; therefore, measures need to be taken for improving truck haulage safety and efficiency. In the absence of any major advancement in the automation technology for the mining sector, this study attempted to eliminate the existing technology lull by developing a novel DeepHaul framework for inducing smartness and intelligence within any dump truck, using the advanced algorithmic knowledge of artificial intelligence and machine learning. The DeepHaul framework consisted of two major components: first, inducing an object recognition ability by using deep learning methodology, for any dump truck. Experiments were conducted with different deep learning architectures, different training batch size ranges, and various image sizes for developing an optimum deep learning model with state-of-the-art performance for the haul truck. The second component consisted of the steering action decision making ability for the dump truck based on the given state of the haulage route. A reinforcement learning-based algorithm was designed, implemented, and tested for achieving the aforementioned objective. The algorithm exhibited an accuracy of 100% regarding safety and an average accuracy score of more than 97% regarding haulage efficiency. With the implementation of this DeepHaul framework, the existing autonomous haulage truck control technology can be greatly enhanced by inducing intelligence and smartness within the haul trucks, which would result in improved safety, efficiency, and the effectiveness of mining operations.


Mining Engineering

Keywords and Phrases

Autonomous truck; Deep learning; Haul road; Haul truck; Haulage efficiency; Surface mining

International Standard Serial Number (ISSN)

2192-6352; 2192-6360

Document Type

Article - Journal

Document Version


File Type





© 2021 Springer, All rights reserved.

Publication Date

10 Feb 2021