Enhancing Robotic Perception for Autonomous Roof Bolting Using an Event Based Machine Learning Framework - SME Annual Meeting 2024

Underground mine roof bolting is a crucial operation for miners’ safety and mine sustainability. Since roof bolting is a manual or human-supervised operation, miners’ safety is at risk due to dust or rock falls. Traditional machine learning algorithms have shown limitations to detecting drillable areas, mainly due to harsh lighting conditions. The authors propose an adaptive deep-learning framework for autonomous roof bolting. The proposed framework is based on implementing a binary semantic segmentation algorithm on color images to classify pixels that belong to rock from those that belong to non-rock. Significantly, the proposed framework implements deep learning semantic segmentation on images from traditional and neuromorphic vision sensors in underground mines. The performance of the proposed model shows an impressive accuracy level of at least 98% at a low number of training epochs with smooth learning curves. The high accuracy enables the implementation of autonomous roof bolting, greatly improving miners’ safety and operational efficiency while reducing human exposure to safety hazards. This research will advance the use of deep learning in mining automation and has the potential to revolutionize the traditional mining industry.