Multiphase Scouting Control of an Agricultural Field Robot With Reachability Analyses

Abstract

Accurate path scouting control of an autonomous agricultural robot is substantially influenced by terrain variability, field patterns, and uncertainties in sensed information. Based on conventional farming techniques, the targeted test crop of strawberries grows in semi-structured environments. Thus in this study, the proposed scouting control architecture comprises of three distinct portions and in each portion different sensors are used. Based on range finder (RF) information, the first region uses a proportional-integral-derivative (PID) controller with logic steps to account for undesirable pop-up events. In the other two portions, vision-based robust controllers are developed, in which a new bound is derived for the focal length uncertainty in vision. Stabilities of the controllers are proven and the reachabilities are analyzed to guarantee that the final state of each portion is within a desired initial region of the next portion controller. The proposed multiphase scouting control is successfully validated for our custom-designed robot in a commercial strawberry farm.

Minimum‐time row transition control of a vision‐guided agricultural robot

“…The robot transits between rows at headlands, only relying on RGB cameras. As discussed earlier, our previous row transition (or crossbed) controllers follow a nonoptimal square trajectory (Freese & Xu, 2019;. Additionally, sands in a typical strawberry field are loose and tend to build up under the wheels when the robot is turning, causing both slippage and inconsistent turns.…”

“…As discussed in Freese and Xu (2019) and , navigation in a typical strawberry field consist of two phases, over-bed and cross-bed. Controllers corresponding to these two phases serve distinctive purposes using feedback information from different sets of sensors.…”

“…For a large portion of farmlands that have row crops, robot motion control normally consists of row‐wise translations (or over‐bed) and row transitions at headland (or cross‐bed) (Freese & Xu, 2019; Underwood et al, 2017). There is a rich literature on row‐wise translation in different types of farmlands (Ball et al, 2016; de Paula Veronese et al, 2016; Dong et al, 2011; Freese & Xu, 2019; Higuti et al, 2019; Le et al, 2020; Nagasaka et al, 2009; Schwendner et al, 2014; Underwood et al, 2015, 2017; van Henten et al, 2002; Winterhalter et al, 2021; Xiong et al, 2020); nevertheless, studies on row transition controls are limited (Ball et al, 2016; Freese & Xu, 2019; Le et al, 2020; Li et al, 2020; Nagasaka et al, 2009; Underwood et al, 2017; Xiong et al, 2020). In this paper, we present a real‐time subspace optimal control method that significantly improves the row transition performance of a field robot in a strawberry field.…”

“…Hague and Tillett (1996) proposes to use vision‐based sensing techniques to localize the robot with respect to crop rows. Our previous work (Freese & Xu, 2019) uses the color pattern of a typical row‐based field to transit between rows, which is not robust in varying lighting conditions and often results in alignment errors when the rows are not perfectly straight. In Li et al (2020), the visual marker‐based cross‐bed motion is studied, which can mitigate the influence of changing lighting conditions.…”

“…In Li et al (2020), the visual marker‐based cross‐bed motion is studied, which can mitigate the influence of changing lighting conditions. These two existing control schemes (Freese & Xu, 2019; Li et al, 2020) are not optimal. It is worth noting that visual fiducial systems that provide measurements by visual markers have been widely used in robotics (Fiala, 2005; Kalaitzakis et al, 2021; Olson, 2011).…”