Data Augmentation for Continuous Excavation Trajectories Based on DGAN

Artificial intelligence technology represented by machine learning is shining brightly in the industrial field. A key challenge for excavator automation in the construction industry is the lack of high-quality open source continuous excavation datasets. We use an advanced generative adversarial network called DopplerGANer to enhance real-world continuous excavation trajectories. This results in a synthetic yet realistic dataset for model training. Our analysis involves the use of statistical techniques to verify the similarity and diversity of the synthetic data to the original data. In addition, we use standard operating evaluation metrics to evaluate real trajectories and synthetic trajectories in a simulation environment. The results show that the synthetic trajectories not only closely mimic the original data but also provide an improved range of variation. This research lays a solid foundation for the development of reinforcement learning applications in future excavator automation.