Building Autonomous Vehicles (and Importance of Domain Subjects)

Hello Everyone!

In the spirit of yesterday's class session on Simulators and Autonomous Systems in Robot Autonomy Course, I would like to share details about the Autonomous Weeder we developed last semester. This project was a collaborative effort with some sophomores, and here we highlight how we can combine domain knowledge with robotics to build new solutions.

Side View

Front View

Side and front view of the autonomous weeder

Overview of the Autonomous Weeder Project

The weeder was designed to address one of the most labor-intensive tasks in agriculture: weed control. Traditional methods, such as manual weeding or heavy reliance on chemicals, are inefficient and unsustainable, especially in terrain farming, where slopes and varying soil conditions make the process even more challenging. Our goal was to create a solution capable of autonomously navigating such environments while minimizing the environmental impact.

Key Features

 1. Terrain-Adaptability and Drive System

The weeder was built to work in diverse terrains—whether rocky, muddy, or sandy. To achieve this, we used a belt traction system for stability on steep slopes, to allow the weeder to move through uneven land. We wanted to have a suspension design for better traction, but it was not possible in the given time frame. We used EV Scooter hub motors to develop the drive system, which was powerful enough for our needs. The minimum torque requirement was calculated based on the slope of the terrain we chose for reference and using common values of friction coefficient.

2. Blade Design & Engineering Analysis

This aspect of the project required a strong interdisciplinary approach and was the most exciting part.

• Mechanical Engineering played a critical role in calculating key blade parameters, such as material selection (galvanized iron), geometry, and sharpness. Detailed stress analysis and structural simulations were performed to ensure that the blades could withstand forces encountered in terrain farming, such as rocks and compacted soil. 

• Agricultural Knowledge was vital in determining how the blade would interact with different soil types, crops, and weed varieties. This allowed us to fine-tune the blade design for both inter-row and intra-row weeding, to ensure minimal crop damage. 

  

  
  
  

3. Autonomous System

• Using a differential drive mechanism, the weeder followed predefined waypoints to cover the entire field. We built the low-level controller for the differential drive and used ROS for the entire pipeline. GPS and IMU sensors assisted in localization and waypoint tracking.

  
  

4. Stress Analysis

• The entire frame underwent a detailed stress analysis using the ANSYS Static Structural Module. This analysis helped us evaluate the distribution of forces across the frame when subjected to the weight of the components, the dynamic loads from the motors and traction system, and the vibrations caused by uneven ground. The results allowed us to identify weak points in the structure and reinforce those areas with additional support.

Equivalent Stress Analysis of X-frame

5. Equivalent Stress Analysis of X-frame with Sheet

• Equivalent Stress simulations revealed areas of high-stress concentration, which we addressed by adding reinforcement at critical points to ensure the frame could handle both operational loads and environmental stresses

6. Weed Detection and Removal

• The YOLOv8-based system for weed detection was trained on agricultural datasets to identify weeds in real time. The system provides bounding boxes around the weeds, which the autonomous stack uses to guide blade movements.

7. Electronic Architecture

The power and control system was designed to manage both rover movement and blade actuation. Stepper motors provided precise control, which was especially crucial for intra-row and inter-row weeding tasks.

High Level Electronic Architecture

I wouldn't say the build was perfect. There were other nuances like synchronization of object detection, weed cutting, and speed of the vehicle, which we manually tuned but didn't obtain a good result. We had problems with the gear systems, especially properly aligning them (the chain would become too tight or loose) while fabricating. In the end, we just patched it somehow.

Impact and Future Prospects

The autonomous weeder has the potential to transform agriculture by reducing manual labor and reliance on harmful chemicals. This system could be integrated with other farming operations, such as seeding and plowing, creating a fully automated agricultural system.

The development of this project was a great learning experience in blending different fields: Mechanical Engineering, Computer Science, and Agriculture. Each discipline played a critical role in the successful creation of the weeder. This project exemplifies what can be achieved when knowledge from multiple domains comes together to solve real-world challenges.

A common observation I’ve made among students in IIT Kharagpur is that many tend to remain within the boundaries of their chosen field, often missing out on the opportunities that interdisciplinary knowledge can provide.

For instance:

• CS students may not realize that, with some knowledge of Aerospace Engineering, they could build high-fidelity simulators for rockets, jets, or drones or even create digital twins for complex systems. (I might share more about one of the simulators I developed in a future post. 😎)

• Mechanical Engineering students can use their coding skills to write Computational Fluid Dynamics (CFD) solvers or Finite Element Method (FEM) solvers, maybe, let's say to design and optimize structures for serial manipulators.