Sonar-based aid for the execution of headland turns by a vineyard robot

Abstract. The EU-funded VineRobot project aims at designing, developing, and deploying a practical agricultural robot endowed with non-invasive biosensors to map canopy growth in vines and red grape maturity. After two years of development, autonomous navigation inside the vine rows achieved satisfactory results with the only use of a stereoscopic camera, which allows for the real-time processing of the 3D scene ahead of the robot. However, turning at the headlands entails a challenging scenario for a medium-size robot that must dive into the field every two rows without hesitation, being the terrain conditions unpredictable and close-up perception out of reach for the bifocal stereo camera. Even though crop mapping only occurs inside the field, and the majority of navigation time always elapses when moving along the rows, a safe and reliable maneuver at the headlands is necessary to grant the general acceptance of the robot. To strengthen the navigation skills of the robot while maintaining a moderate level of complexity, the safety bumper was endowed with a network of ultrasonic devices that expanded the perceptive range of the robot for the critical 1-m strip at its front. This paper explains the configuration of this smart bumper, its workings, and the behavior of the robot when making U-turns at the headlands during preliminary tests conducted in French vineyards in the spring of 2015. A multi-stage approach yielded stable results in the delicate process of changing rows.