Structure tracking strategies in different agricultural configurations

Abstract

Navigating in an off-road environment is a challenging task. Indeed, robots need to recognize and detect significant elements on the path in order to navigate and perform required tasks in these environments safely. In this paper, it is proposed to study two different structure tracking strategies, that depend on the disposition of a 2D LiDAR perception solution that allows for fulfilling different agricultural tasks. The first strategy positions the LiDAR horizontally w.r.t. to the soil which permits extracting the outer contour of the structure to be followed by taking advantage of an extended vision provided by this disposition. This allows anticipation and obstacle avoidance capabilities. The second approach positions the LiDAR inclined toward the soil, allowing the recognition of the shape of a structure of interest thanks to the 3D perspective of the environment provided by the sensor layout. The 3D perspective provides knowledge on the presence or absence of vegetation. The problem associated with accurate control for mobile robots following a structure is addressed thanks to a backstepping control. In this control strategy, the longitudinal and lateral controls are computed independently and thus the regulation of the position and the orientation of the robot are separated. It is used in order to control a 4-wheel-steered mobile robot in an experimental field. The efficiency of both approaches is then investigated through full-scale experiments in various conditions.