Waypoint Generation in Row-Based Crops with Deep Learning and Contrastive Clustering

The development of precision agriculture has gradually introduced automation in the agricultural process to support and rationalize all activities related field management. In particular, service robotics plays a predominant role this evolution by deploying autonomous agents able navigate fields while executing different tasks without need for human intervention, such as monitoring, spraying harvesting. context, global path planning is first necessary step every robotic mission ensures that navigation performed efficiently with complete coverage. paper, we propose learning-based approach tackle waypoint generation row-based crops, starting from top-view map region-of-interest. We present novel methodology clustering based on contrastive loss, project points separable latent space. proposed deep neural network can simultaneously predict position cluster assignment two specialized heads single forward pass. extensive experimentation simulated real-world images demonstrates effectively solves problem both straight curved overcoming limitations previous state-of-the-art methodologies.