← Back to Projects

ASABE Agricultural Robotics Challenge Robot '26

Fully autonomous cornfield robot — line following, plant classification, and precision removal

ROS2 YOLO Python Jetson Nano Orin Arduino Mega Mecanum Wheels
ASABE Robot in action

Why

Built for the ASABE Student Robotics Challenge 2026 — Standard Division. The task: traverse a 5-lane simulated cornfield, identify bad (yellow) corn plants among good (green) ones, and remove only the bad ones using a precision actuator — all autonomously, as fast as possible.

As VP of the MSU Precision Agriculture Robotics Club (PARC), I led the team across mechanical, electrical, and software subteams from initial design through competition-ready testing. The result is a fully autonomous robot built on a custom 3D-printed chassis, Mecanum wheels for omnidirectional movement, and a ROS2 pub/sub architecture tying navigation, vision, and actuation together.

With

📡 Sense

  • 2× HBV CAM V2101 V11
    Front camera: line following
    Side camera: row transition

💻 Think

  • Jetson Nano Orin
    YOLO detection · ROS2 nodes · camera processing
  • Arduino Mega
    5 motor controllers via GPIO

⚙️ Act

  • 4× FIT0441 Gearmotors
    60mm Mecanum wheels — omnidirectional drive
  • Plant Removal System
    Lateral slide + rotational end effector lifts bad plants off magnetic base

Control Flow

  1. Follow — detect_node reads front camera, publishes line state; drive_node runs PID to minimize pixel tracking error
  2. Detect — YOLO model classifies each plant base as Green Corn, Yellow Corn, or empty
  3. Decide — on Yellow Corn detection, drive_node halts robot and triggers actuation sequence
  4. Remove — actuator slides laterally, aligns prongs, rotates to lift bad plant off magnetic base, resets
  5. Transition — at row end, detect_node switches to side camera to navigate to the next lane

Wrinkles

Live Detection vs. Training Accuracy

Challenge — YOLO trained on 255 still images under controlled lighting. Live camera conditions introduced lighting variance that dropped precision/recall below training metrics.

Solution — Identified the gap between static image training and live camera inference. Planned re-training with augmented lighting conditions for future iterations; adjusted confidence thresholds for the competition environment.

Actuation Alignment Tolerance

Challenge — The end effector prongs had to align precisely with the plant base center — too far off and the plant wouldn't lift cleanly off its magnet.

Solution — Simplified the actuation system to a single-axis movement (horizontal + one rotational axis), minimizing degrees of freedom and error accumulation compared to a multi-DOF robotic arm. Tuned stop position using bounding box center coordinates from the vision system.

Wins

Tested on a full-scale replica of the ASABE competition arena.

  • Fastest run: 62.63 seconds end-to-end
  • Average time per row: 11.37 seconds
  • YOLO mAP: 0.991 overall — 0.995 Base · 0.984 Green · 0.995 Yellow
  • Plant ID scoring: projected 100% singles · 90% doubles
  • Navigation: reliable line following across all documented test runs

Read the Full Design Report →

Which

🔧 Mechanical

  • Custom chassis — Fusion 360, Bambu Labs P2S, PLA+
  • Square form factor housing all electronics + batteries
  • 4× motor mounts for Mecanum wheels
  • Camera mounts + center wire pass-through
  • Actuator housing on underside

⚡ Hardware

  • Jetson Nano Orin — main compute
  • Arduino Mega — motor control
  • 4× FIT0441 DC Gearmotors + 60mm Mecanum wheels
  • 2× HBV CAM V2101 V11 cameras
  • 12V 5200mAh LiPo battery (>1 hr runtime)
  • Plant removal system: lateral slide + rotational end effector

💻 Software

  • ROS2 — pub/sub node architecture
  • drive_node — PID line following
  • detect_node — camera switching + color detection
  • YOLO — plant classification (Roboflow, 255 images)
  • Python — all nodes and vision pipeline

Where

Competed at the ASABE Annual International Meeting 2026 in Indianapolis, Indiana — the premier conference for agricultural, biological, and food engineers. The robot competed in the Student Robotics Challenge Standard Division against university teams from across the country.