Impact of autonomous robotic weeding on soil physical properties across a soil texture gradient

The market for autonomous agricultural robots is rapidly developing, with models differing in size, weight, and field applications. Operations range from tilling and seeding to pest and weed control and harvesting. In the transition from simplified to more diversified cropping systems, agricultural robots could support the implementation of labour-intensive, spatially diversified cropping systems by operating autonomously in the field. They may also enhance soil health by replacing herbicide applications with mechanical weed control. However, there are very few studies that quantify the effects of (lightweight) agricultural robots performing mechanical weed control on soil physical properties and overall soil health.

In this study, we aim to assess the impact of mechanical weed control conducted by the autonomous field robot FarmDroid FD20 (FARMDROID APS, Denmark), a solar-powered robot with a working width of 3.5 m and a weight of approximately 1250 kg, on soil physical parameters. It can conduct seeding and subsequently, inter- as well as intra-row harrowing and hoeing with different tools, like blades, cutting knives, disks and wires, equipped to the robot. The weeding depth is approximately 4 cm.

The assessed parameters include bulk density and penetration resistance across three contrasting sites in Germany, sampled at two time points in 2025: before tillage in spring and after several weeding interventions in summer.

The sites span approximately 545 – 830 mm of annual precipitation and differ strongly in soil texture, ranging from sand in eastern Germany to loess-derived silt in central Germany to clay-rich soils in southeast Germany. In block designs with three to four replications, either cultivated with maize (Zea mays) or sugar beet (Beta vulgaris), we assessed the absence of mechanical weed control (herbicide application) compared to mechanical weeding with the FD20. In one of the experiments, two intensity levels of mechanical weeding were also included.

We determined bulk density in trafficked and non-trafficked areas by collecting a total of 570 soil cores (100 cm³) at three depth intervals: 2 – 7, 11 – 16, and 20 – 25 cm. Dried bulk density samples were sieved and corrected for stone content. Moreover, penetration resistance was measured in a transect in the topsoil using a penetrometer (Penetrologger, Royal Eijkelkamp, The Netherlands).

Preliminary bulk density results indicate that robotic weed control and its traffic mainly affected the upper 0 – 7 cm, while effects diminish at greater depths. The analysis of penetration resistance data is still ongoing and will complement the assessment of soil physical responses to robotic weed control.