Toward Precise Robotic Weed Flaming Using a Mobile Manipulator with a Flamethrower

TL;DR

This work tackles precise robotic weed flaming by equipping a mobile manipulator with a flamethrower and developing a real-time flame coverage model. It introduces a center-arc flame representation with a Gaussian cross-section and a two-view flame estimation pipeline that fuses infrared imagery with the nozzle pose to predict flame boundaries. The approach demonstrates robust performance, achieving an mAP above 76% in real-time predictions and high field execution accuracy, including 94.4% precision and a 6.71 cm center offset in raised-bed tests. The contributions enable environmentally friendly, contact-free weed control with practical deployment potential, and lay groundwork for future multi-weed dynamic coverage and extended removal modalities.

Abstract

Robotic weed flaming is a new and environmentally friendly approach to weed removal in the agricultural field. Using a mobile manipulator equipped with a flamethrower, we design a new system and algorithm to enable effective weed flaming, which requires robotic manipulation with a soft and deformable end effector, as the thermal coverage of the flame is affected by dynamic or unknown environmental factors such as gravity, wind, atmospheric pressure, fuel tank pressure, and pose of the nozzle. System development includes overall design, hardware integration, and software pipeline. To enable precise weed removal, the greatest challenge is to detect and predict dynamic flame coverage in real time before motion planning, which is quite different from a conventional rigid gripper in grasping or a spray gun in painting. Based on the images from two onboard infrared cameras and the pose information of the flamethrower nozzle on a mobile manipulator, we propose a new dynamic flame coverage model. The flame model uses a center-arc curve with a Gaussian cross-section model to describe the flame coverage in real time. The experiments have demonstrated the working system and shown that our model and algorithm can achieve a mean average precision (mAP) of more than 76\% in the reprojected images during online prediction.

Figures (7)

• Figure 1: Robotic weed flaming system design and main components.
• Figure 2: Illustration of the robot coordinate systems with color-coded axes.
• Figure 3: Software diagram. The software system is partitioned into two components with the shaded blocks control Spot body motion and the rest directs the manipulator motion for weed flaming.
• Figure 4: (a) Illustration of flame projection. The projected thermal images in the first and second views are shown in the short and long red boxes, respectively. Using the center line (black dashed line), each backprojected ray from $\{C\}$ (blue dotted lines) determines a flame surface point (red stars), while the depth ambiguity in rays from $\{C'\}$ (red dotted lines) blocks the surface point estimation. (b) Illustration of the flame center arc model. (c) Illustration of the flame cross-section width function $w(l)$. The red line shows the fitted $w(l)$ and the black dots are the measurements
• Figure 5: (a) and (b) show the original thermal images and the reprojected images under light wind and strong wind, respectively. The red area is from the circular arc model, the white area is the thresholded region and the bule outline is from the straight line model. (c) and (d) show the corresponding reconstructed 3D flame using the circular arc model.

Theorems & Definitions (1)

• Definition 1: Flame Estimation