Deep Learning-Based Autonomous Excavation: A Bucket-Trajectory Planning Algorithm

Abstract

Increased safety risks and the difficulties of training excavator operators, combined with manpower shortages, have led to an increased demand for machine automation. This study applies a long short-term memory algorithm to automate a bucket tip-trajectory planning artificial intelligence (AI) system. Unlike other autonomous excavation techniques, our approach performs bucket-trajectory planning without prior knowledge of the nonlinear bucket–soil interaction dynamics during excavation, which require the parameters to be precisely adjusted via heuristic analysis of their correlations. Using data acquired from the excavations of excavator experts, this method uses three-dimensional point cloud of terrain and bucket motion data of excavation process to train and apply the AI modules. Especially, we transform the point cloud, which comprises a massive number of points and entails a considerable computation complexity, into much fewer values, which are enough to represent the three-dimensional shape of the target terrain. To ensure prevention of collisions with underground obstacles along a given excavation path, a collision avoidance algorithm, based on continuous pressure monitoring in the excavator’s hydraulic cylinder, is applied. Comparison experiments reveal that the bucket-tip trajectory planning AI system with collision avoidance algorithm generates a traceable trajectory for the machine controller, equipped in an excavator, and yields the desired excavation volume and lead time without collision, regardless of the topographic changes caused by successive excavations.

Keywords

• Automated excavation
• bucket-tip trajectory planning
• collision avoidance algorithm
• field robots
• long short-term memory
• point cloud data