Leveraging Behavior Trees for Hybrid Autonomous Navigation in Seasonal Agricultural Environments

Autonomous navigation is essential for the successful integration of mobile robots in agricultural operations. In structured fields, where permanent crops are usually disposed in row patterns, perception-based navigation is typically used for achieving safe and efficient in-row operation, whereas map-based navigation and other techniques are applied for transitioning between rows and from/to the robot base station. However, execution and coordination of different strategies has been mostly achieved using finite state machines or rule-based implementations, limiting orchestration of complex behaviors and scalability. This work presents a modular reasoning architecture that leverages behavior trees and a topological representation of the environment for deploying agricultural robots, switching between operation modes (perception or map-based) according to their topological state and goal, and embedding recovery behaviors in the event of failure. The system has been validated with different robotic platforms (mobile robot and retrofitted tractor) and large scale pilots (apple orchards and table grape vineyards), resulting in successful autonomous spraying demonstrations.