SCI Release! The latest research results of Sany unmanned excavation technology

It is an international, peer-reviewed, open access journal focused on the field of applied natural sciences. Published online by MDPI and issued twice a month. Rated Q1 by JCR in both Engineering (Multidisciplinary) and General Engineering.

On August 8, 2024, by Sany Heavy MachinerySmall diggingThe paper entitled "E-GTN: Advanced Terrain Sensing Framework for Enhancing Intelligent Decision Making of Excavators" was officially released by the team of Dr. Gao Le, Vice Dean of the Institute, and the team of Professor Hu Jie of Shanghai Jiao Tong University.

Introduction to the paper

In the face of the complex and changeable mining environment, a multi-modal large model technology, the E-GTN framework, jointly developed by Dr. Gao Le's team and Professor Hu Jie's team of Shanghai Jiao Tong University, provides a theoretical solution to the existing challenges of unmanned mining.

Traditional unmanned mining technology relies on expert experience and dynamic programming, but it has limitations in efficiency and generalization. The E-GTN framework uses terrain feature extraction and multi-modal fusion technology, based on reinforcement learning and large model architecture, to significantly improve the decision-making ability and environmental adaptability of unmanned mining. The framework includes three parts: terrain information processing, terrain feature extraction and decision-making.

The E-GTN framework integrates multi-modal fusion technology, and realizes high-precision 3D reconstruction of the excavation environment by combining multiple sensor data, providing richer and more accurate terrain information for unmanned excavators. In addition, the research team proposed a grid mining environment perception technology based on convolutional networks, and the custom-designed model GridNet can extract significant terrain features.

The decision-making process is modeled as a deep reinforcement learning (DRL) algorithm based on large models for mining tasks. This technology not only responds to environmental changes in real time, but also has intelligent decision-making capabilities, enabling unmanned excavators to make efficient and accurate excavation operations in various terrain conditions.

The launch of the E-GTN framework provides a new technical exploration for the automation and intelligent development of the mining industry. With the continuous maturity of large-scale model technology and the expansion of its application scope, it will bring more accurate, safe and efficient unmanned operations.