What is it about?
A farm robot can follow crop rows using a single camera, steering by what it sees (image-based visual servoing, IBVS). The standard version, though, oscillates and can drift off the row when the robot drives fast, the terrain is uneven, or the camera is only roughly calibrated. Robust controllers fix that instability, but the usual ones rely on rapid on/off switching that produces "chattering" — a high-frequency jitter that stresses motors and can shake the vehicle. This work uses the super-twisting algorithm, a second-order sliding-mode control, to add the robustness term: it keeps the disturbance-rejecting power while smoothing that jitter away. The controller works in two phases — a column controller to reach and turn onto a row, and a row controller to stay centred between rows. We tested it in realistic 3D simulations (ROS-Gazebo) with a differential-drive robot driving at 1.5 m/s, with 20% camera-calibration error plus noise and unmodelled dynamics, and compared three strategies: the classic IBVS, a first-order robust controller (unit vector control), and our second-order super-twisting approach. The super-twisting controller tracked the row most accurately and produced the smoothest, lowest-effort steering of the three.
Featured Image
Photo by James Baltz on Unsplash
Why is it important?
The contribution is about getting robustness without the usual penalty. Plain robust controllers reject disturbances by switching discontinuously, which causes chattering that wears actuators and can excite unwanted vibrations; the super-twisting algorithm — a second-order sliding mode — preserves the robustness while attenuating that chattering, giving continuous, smooth control. In the head-to-head comparison the super-twisting approach beat both the classic IBVS and the first-order robust method on every metric: it more than halved the classic method's row-tracking error and produced the lowest control-effort variation of the three (roughly an order of magnitude below the classic controller). All of this runs on a single, inexpensive monocular camera — no GPS-RTK, LiDAR or stereo rig — and the stability is backed by a Lyapunov analysis rather than simulation alone. The practical upshot is smoother, gentler navigation that is kinder to the hardware and the crop, uses less energy, and holds up at the higher driving speeds real field work demands.
Perspectives
What I find most interesting here is the trade-off at the heart of robust control: discontinuous laws give you strong disturbance rejection but at the cost of chattering, and the super-twisting algorithm is an elegant way to keep the former while taming the latter. Putting first-order and second-order sliding modes side by side on the same agricultural problem made that contrast concrete — and the second-order approach came out ahead on accuracy and smoothness, which is not something you can always take for granted. For me, smoothness is not a cosmetic detail: it is what protects the motors, spares the plants, and ultimately decides whether a controller is fit to leave the simulator. The directions I am keenest to pursue are validating this on a physical robot in the field, extending it to more sharply curved rows, and studying how far the chattering-versus-robustness balance can be pushed before it costs tracking performance.
Prof. Dr. Eduardo Costa da Silva
Pontificia Universidade Catolica do Rio de Janeiro
Read the Original
This page is a summary of: Vision-based Autonomous Crop Row Navigation for Wheeled Mobile Robots using Super-twisting Sliding Mode Control, August 2021, Institute of Electrical & Electronics Engineers (IEEE),
DOI: 10.1109/ecmr50962.2021.9568819.
You can read the full text:
Contributors
The following have contributed to this page
