Skip to content

amanavvk/farmbot

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Modular autonomous mobile robot for precision spray of Pesticides

Ubuntu ROS C++

TODO

  • CAD Design of robot (Must be optimized for manufacturing)
  • Program localisation method using GPS, IMU (and Vision if required)
  • Research about architecture of CNN to be used..
  • Train for weed detection purpose
  • define and solve lane tracking problem as input image and output Lane msg (to be modified)
  • deploy CNN in Jetson / Pi for weed detection purpose using TensorRT
  • Nozzle position planning algorithm

Description

A farming robot with automatic navigation, weed detection system, designed for navigating between crops, across farms and providing aid in multiple operations required by farmers, specific in automatic weed spraying system. Automatic weed detection and localization is achieved with deep neural network running on NVIDIA jetson nano, then planning sprayer path and spraying precisely on weeds. It is developed using ROS as its core, leveraging the open-source community to each components which makes system modular, efficient and scalable. GNSS and RTK based approach is used for localization, and for increasing accuracy, sensor fusion method is used for more accurate state estimation. Localization is then used with vision system to guide through crop lanes and then deep neural network running parallel to lane tracker program detects weeds with locations and plans trajectory of nozzle and spraying time stamp and controller will execute planned trajectory.

Objectives

  1. Reduce amount of chemical (weedicide) usage in crops using automatic high precision weed detection (using CNN) and low-cost spraying system.
  2. Develop RTK based automatic navigation system for accurate positioning and motion across different lanes of crop field.
  3. Develop base platform with modular system for further improvisation in multi - functionality of robot (future work).

Major Scientific fields of Interest

Such autonomous precise weedicide spraying robots are available in market but are too expensive because they run on heavy Deep Neural Networks, which are GPU- intensive and it’s not really needed because the risk factor is less so the trade-off of accuracy and speed can be favored more towards speed and therefore we would be using a convolutional network with lesser number of layers. Since the number of classes here are less (say, crops, weeds and lane), A narrow neural network would still work, which would even work on a common small-scale robotic system viz, Raspberry Pi and NVIDIA Jetson Nano.

Approach

  • Vision and GNNS + RTK based autonomous navigation of vehicle.
  • Development of ecosystem of modular systems using robot operating system (ROS) for farming application.
  • Design of vehicle with certain payload capacity and mobility. Vision based weed and crop detection, classification, localization and precisely spraying over weeds.
  • Robot operating system will interface with micro-controllers to control vehicle trajectory, sprayer nozzle position and get sensor data.
  • Localization of vehicle using GNSS + RTK based system and implementing sensor fusion method to improve state estimation of vehicle.

FLOWCHART :

3D Model of the Farmbot

References

  • Meshram, A. T., Vanalkar, A. V., Kalambe, K. B., & Badar, A. M. (2022). Pesticide Spraying Robot for Precision Agriculture: A Categorical Literature Review and Future Trends. Journal of Field Robotics, 39, 153–171. LINK

Contributers

Aman Vyas, Bhavik Kasundara and Avdhesh Kumar