Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Dhruv2012/Autonomous-Farm-Robot/llms.txt

Use this file to discover all available pages before exploring further.

AGRIBOT is an open-source autonomous agricultural robot designed to eliminate weeds on a real-time basis without any human involvement. Built as an undergraduate project at the Sardar Vallabhbhai National Institute of Technology (SVNIT) under the guidance of TEQIP-3 and the ECE Department, AGRIBOT combines mechanical design, embedded systems, robotic simulation, and deep learning into a single end-to-end system. By automating weed removal, AGRIBOT aims to reduce dependency on manual labor, improve crop yield quality, and offer a scalable platform that can be extended to other farming tasks such as ploughing and harvesting — ultimately contributing to agricultural GDP and farmer welfare.

​
What AGRIBOT Does

AGRIBOT is built around two core technical pillars that work in tandem: Autonomous Navigation and Crop-Weed Classification. The navigation subsystem drives the robot through crop rows using GPS and magnetometer data fused with Kalman filtering, while the classification subsystem uses a convolutional neural network to segment every camera frame into crop, weed, and soil regions — triggering targeted weed removal without disturbing surrounding plants.

Autonomous Navigation

GPS + IMU-guided path planning with Moving Median and Kalman filters for noise reduction. Implements skid-steer drive control in Gazebo simulation.

Crop-Weed Classification

CNN-based semantic segmentation using UNet and Bonnet architectures trained on the CWFID and Bonn sugar-beet datasets. Bonnet runs at ~2.5 fps for real-time inference.

System Architecture

Modular ROS Catkin workspace with four packages: robot description (URDF), controllers, autonomous drive nodes, and Gazebo simulation worlds.

Quickstart

Set up the ROS workspace, launch the Gazebo simulation, and run the navigation stack and classification pipeline from scratch.
AGRIBOT was developed and tested primarily in Gazebo simulation due to restrictions imposed by the COVID-19 pandemic, which prevented full deployment on the physical robot. However, the navigation algorithms are validated in simulation with realistic sensor noise, and the crop-weed classification models are validated directly on real farm images captured from sugarbeet fields and surrounding farms near SVNIT.

​
Published Research

AGRIBOT’s design and algorithms are documented in a peer-reviewed paper presented at the AVES 2021 Conference and subsequently published by Springer: “Design of an Autonomous Agriculture Robot for Real Time Weed Detection using CNN” Dhruv Patel, Meet Gandhi, Shankaranarayanan H., and Anand Darji. arXiv:2211.12077 · Springer Chapter The paper covers the full system design, including embedded and mechanical architecture, URDF-based robot modeling and Gazebo simulation, GPS/magnetometer path planning, and deep learning crop-weed classification results on both the CWFID and Bonn datasets.

​
Cite This Work

If you use AGRIBOT’s code, models, or methodology, please cite the paper using the following BibTeX entry: 
@misc{https://doi.org/10.48550/arxiv.2211.12077,
  doi = {10.48550/ARXIV.2211.12077},
  url = {https://arxiv.org/abs/2211.12077},
  author = {Patel, Dhruv and Gandhi, Meet and H., Shankaranarayanan and Darji, Anand D.},
  title = {Design of an Autonomous Agriculture Robot for Real Time Weed Detection using CNN},
  publisher = {arXiv},
  year = {2022},
  copyright = {Creative Commons Attribution 4.0 International}
}

​
Project Team

AGRIBOT was conceived, designed, and built by a team of undergraduate students and their faculty guide at SVNIT Surat.

Dr. Anand Darji

Project Guide — Head & Associate Professor, Electronics & Communication Engineering Department, SVNIT.

Development Team

Dhruv Patel (U16EC053), Meet Gandhi (U16EC056), Shankaranarayanan H. (U16EC074) — all from SVNIT, Surat.
Special thanks to Mr. Mahesh Birajdar (U16ME084) and Team DRISHTI at SVNIT for their contributions to the SolidWorks design and physical manufacturing of the AGRIBOT chassis.

​
Key Capabilities at a Glance

  • GPS + Magnetometer Navigation — waypoint following using Mercator projection coordinate conversion and atan2-based heading computation
  • Sensor Noise Reduction — Moving Median filter and single-dimension Kalman filter applied to heading data from the MPU-9265 magnetometer
  • Semantic Segmentation — Bonnet CNN architecture with ~100× fewer parameters than UNet, enabling real-time inference on an Nvidia Jetson Nano
  • Dual Dataset Validation — trained and evaluated on CWFID and Bonn Sugar-Beet 2016 datasets
  • Full Gazebo Simulation — complete sensor suite (GPS, IMU, magnetometer, camera) simulated with realistic Gaussian noise and drift parameters
  • Open Source — all ROS packages, CNN training scripts, URDF models, and dataset tools are publicly available on GitHub

Build docs developers (and LLMs) love

Get started for freeTalk to us