Shovel point optimization for unmanned loader based on pile reconstruction

Abstract

This study details an advanced shovel point optimization system for unmanned loaders, crucial for efficient shovelling operations. First, the shovel point evaluation index is established with reference to the driver's experience. Second, a novel method for pile profile reconstruction is proposed, utilizing a trained neural network to detect piles and extracting the point cloud using LiDAR and camera fusion. Subsequently, the system employs optimization algorithm to identify the best shovel point. Finally, 62 consecutive working experiments are successfully conducted. The system's performance closely approximates the driver's choices and achieves an average bucket fill factor of 97.7% for four materials. Results demonstrate the proposed method is reliable and efficient and contributes to the development of automated construction machinery.