Published: May 15, 2025

Weed Warden is an automated weed removal robot targeted for residential lawns, designed to remove dandelions using computer vision-based detection and a precision 2-axis Cartesian positioning system. A Raspberry Pi 5 runs a YOLOv11-Pose model to detect weeds, while an STM32 Nucleo F446RE microcontroller controls the YZ positioning system, which aligns a drill-based extraction mechanism for effective weed removal. A 3-wheel, FWD locomotion platform enables efficient spatial navigation, making Weed Warden a chemical-free alternative to traditional lawn maintenance across diverse outdoor environments.

Weed Warden, as a final product, is a weed control solution designed to periodically traverse the user's lawn, aiming to maintain a weed-free appearance and reduce weed regrowth. It achieves this through multiple drill attachments to remove surface-level weeds and disrupt taproots, decreasing the likelihood of weeds returning.

Technical Summary

Weed Removal Mechanism

Drill-based weed removal system enables 100 mm ground penetration for effective taproot removal

2-Axis YZ Cartesian Positioning System

Y-Axis: Linear rails with lead screw system, driven by NEMA 17 stepper motor for precise horizontal movement of drill
Z-Axis: Ball-screw-driven linear actuator for smooth, controlled vertical positioning of drill

Locomotion

3-wheel locomotion platform for spatial mobility of robot
FWD system, powered by brushed DC motors with reduction gearboxes

Compute & CV System

Raspberry Pi 5 for main compute, handles high-level decision-making and image processing from Pi Camera Module 3
YOLOv11-Pose model for dandelion detection
2-axis Cartesian system controlled by STM32 NUCLEO-F44RE microcontroller

The following video showcases Weed Warden's autonomous weed control cycle, performed on a fake lawn setup, with limited drill speed for safety.

full_clear_8xSpeed.mp4

The following video showcases Weed Warden's repeatability. The robot completes two passes of the turf and docks itself in between runs.

timed_double_clear_16xSpeed.mp4

The Team

Motivation & Scope

From The Start

Initial Concept Generation - Weed Removal System

Engineering Design Specification

Initial Design Layout

ME 481 Final Design Layout - (Sept 2024 - Dec 2024)

Camera Frame Assembly

The Team

This project was completed in a group of 4 for our final year design project (FYDP). The following list details the group members and their respective roles:

Ethan Dau: (Project & Electrical Lead) responsible for design of electrical system architecture and electrical fabrication. Led PM tasks.

Logan Hartford: (Technical & Software lead) responsible for developing the software and control system of the robot (main compute, CV-model).

Varrun Vijayanathan (myself): (Mechanical lead) responsible for overall mechanical design of the system from concept generation -> CAD Layout -> Assembly. Primary focus was development of the weed removal system.

Eric Gharghouri: Part of the mechanical design team (concept generation, CAD). Primary focus was development of the locomotion assembly and design verification tasks.

Given my role of Mechanical lead, this webpage mainly highlights and discusses the design of the mechanical system. For full technical details, please refer to the design report(s) at the end of this webpage.

Additionally, if you would like to learn more about the autonomy/software components, please refer to Logan's website.

Motivation & Scope

Throughout the spring and summer months, residential lawns are prone to broadleaf weeds, such as dandelions. Dandelions are invasive and often deemed unappealing, hence many homeowners seek to remove them. One common method of removal involves using herbicides, such as the Trimec formulation, popular for most broadleaf weeds. However, Trimec, like many herbicides, poses a toxicity risk towards humans, pets, birds, fish, and more. An eco-friendlier removal method involves manually extracting the root, typically using a trowel or weeder. However, this requires users to repeatedly bend over and work in awkward positions, which strains the lower back, leading to inflammations and muscle spasms.

Given these issues, we saw an opportunity to introduce robotics and autonomy into the residential weed removal industry. Some existing products include the Dandy DT-101, Tertill or Nexus Robotics. The Dandy DT-101 uses a herbicide-based approach and mimics the mechanical design of common autonomous lawnmowers. The Tertill uses a string trimmer-based approach for weed control, but is limited to working on small vegetable gardens. The last option, provided by Nexus Robotics, is a Zamboni-sized weed removal solution for large farms. Given the different applications and methods of weed removal undertaken by the three products discussed, there is an opportunity to develop a non-herbicide based autonomous robot to weed residential lawns.

This capstone project was intended to be completed within an 8-month time frame, split across two courses (ME 481 & ME 482). Thus, the main focus of the project was the development of a functional prototype of such autonomous weed removal robot, with the full design of the weed removal system being the focus for ME 481 and the development of the locomotion system for ME 482.

From The Start

The following section provides details of the entire design process undertaken to reach the final CAD layout and initial prototype.

Based on existing solutions, we broke down the final goal of creating an autonomous weed removal robot to better manage our goals across the 2 school terms:

Weed Removal System - locating and removing weed
Locomotion Platform - global navigation of the prototype

The weed removal system was the focus for the first term, whereas the locomotion platform was the focus for the latter

Initial Concept Generation - Weed Removal System

Breaking down the weed removal system further, we split it into two sub-systems based on feedback from our faculty advisors:

Weed Removal End Effector - for actual removal of the weed from the lawn
Cartesian Positioning System - for fine position control of the weed removal end effector

Initial concept generation started with determining the optimal weed removal end effector, as the rest of the robot's design largely centered around it. Since herbicides were not a route we wanted to pursue as mentioned earlier, we primarily focused on mechanical removal methods. The 3 options explored included:

Coring Drill Bit - using a coring drill bit to extract a core of soil containing the weed and collect it for effective removal
Regular Drill Bit - using a regular drill bit to drill the root of the weed itself
Claw End Effector - use a solenoid-activated claw, similar to existing hand-held weed removal tools

Based on preliminary force testing for drill-based setups and the claw option (using Fiskars weed removal tool), the drill-based option, with a coring drill bit had the lower force requirements and thus, was deemed the path forward.

After choosing the weed removal end effector, we moved onto the cartesian positioning system, which was responsible for the fine positioning of the weed removal system to give better fidelity to the end effector and ultimately better accuracy. The two main system types considered were CoreXY (typically employed in 3D-printers) and a general Gantry system, for 3-axis control of the weed removal end effector.

After long team discussions, we came to the agreement it was a bit redundant to include a full 3-axis (XYZ) cartesian system, regardless of implementation, given that the potential locomotion system could take care of one axis of motion very well (through forward/backward translation). Therefore, we simplified the cartesian system to the following details:

Cartesian System: 2-axes of end effector motion (YZ)
Locomotion System: X-axis motion

Engineering Design Specification

The table below describes the engineering design specification revised after initial concept generation. Only minor revisions were made to the EDS throughout the term, with all aspects of the robot design (hardware, software, etc) being made to achieve these specifications at the end of the project.

Initial Design Layout

There were many intermediate design steps/decisions taken to reach the initial CAD layout of the weed removal system, so I have highlighted some of them here on this page.

Some initial concept sketches I completed of Weed Warden's frame and cartesian system are shown below. The weed removal system, as a whole, would consist of the bulk of the components and would serve as a foundation of the robot, with the locomotion system being an add-on down the line. The frame was conceptualized to be made from aluminum extrusions, with the Y-axis frame a central part of it. The z-axis would be connected to the y-axis system, with its linear motion taken care of common methods like lead screws, linear motion shaft and stepper motors.

With these initial concept sketches in mind, the next step was to determine the physical sizing of Weed Warden, which would help me better size motion components and design any custom 3D-printed/machined components. To determine physical sizing, I referenced existing autonomous lawn care products, specifically autonomous lawn mowers as they are a matured product category within this space. The table below presents the sizes seen from a sample of products:

Based on the ranges we provided, we agreed upon some target dimensions for our weed removal system, that were below the dimensional constraints in the EDS:

Length: 450 [mm]
Width: 400 [mm]
Height: 450 [mm]

The below image shows a better depiction of the planned height of the robot, along with a height analysis of the drill system to determine the required range-of-motion (ROM) for it to reach the 100 [mm] penetration depth requirement while also clearing the spade bit and collection bit, used for removal of the dirt core from the coring bit and subsequent storage of the removed cores. A legend for the dimensions in the image below is as follows:

A: Ground Clearance Requirement
B: Penetration Depth FR
C: Drill-stack assembly height (based on sum of drill motor, chuck and coring bit)
D: Length of spade bit
E: Assumed clearance between corning bit and spade bit to prevent collision

Having established the high-level technical specifications for the first design, I started working on the initial CAD layout of Weed Warden.

ME 481 Final Design Layout - (Sept 2024 - Dec 2024)

The following images present the first full design iteration of Weed Warden during the ME 481 term. The design was completed and signed off for production by October 25, 2024, with 3 weeks expected for mechanical fabrication, assembly and electrical integration ahead of initial power-up in December 2024.

Note: In all CAD images, purple-colored components are those designed for 3D-printing.

Working Principle(s) of Design:

Lateral motion of the drill end effector is taken care of the y-axis assembly
Vertical motion of the drill end effector is taken care of the z-axis assembly
X-axis motion is taken care of the locomotion system (forward & backward motion) (integrated in ME 482)

The design features the following features/assemblies:

40x40 mm Aluminum Extrusion Frame - chosen for modularity and ease of reconfiguration in the future
Y-Axis Module - contains z-axis linear stage actuator, drill assembly, y-axis drive system
Collection Bin Assembly - contains spade bit and bin. Coring drill will lower itself onto spade bit to eject weed core (Lower Right of 1st image)

The following images depict the y-axis assembly, which enables the YZ motion of the drill end effector.

Y-Axis Design Details:

Dual HIWIN MGN12 linear rails with MGN12H carriage blocks for enhanced precision and compact design (as opposed to linear motion shafts)
Tr8x8 lead screw, radially supported by two 3D-printed bearing blocks to achieve simple-simple end fixity
NEMA 17 motor used to power the y-axis drive through rotation of the Tr8x8 lead screw
3D-printed cable chains for effective wire management during constant axis motion
3D-printed guards to remove pinch hazards and house limit switches

To reduce mechanical complexity and R&D time, the choice was made to use an off-the-shelf 300 [mm] linear stage actuator equipped with a NEMA 23 stepper motor and ball screw linear for the actuation of the drill assembly in the vertical (z) axis.

Analysis

Hand calculations or FEA simulations were completed to ensure our mechanical design choices were evidence-based. I've highlighted below a few of the design analysis I completed throughout the project, with full analysis details found in the design reports linked at the end of this webpage.

To accurately size the lead screw for the y-axis motion system, the optimal root diameter needed to be determined. To do this, the limiting root diameter was calculated based on screw whip (critical speed) and buckling considerations, given that they are both functions of the root diameter.

Screw whip refers to the oscillation of the lead screw when it surpasses its critical speed, Nc, which is the angular velocity that excites the screw’s natural frequency, causing it to resonate continuously which can lead to failure.
Buckling is the deformation of the screw due to the excessive compressive load applied to it, where the critical load is the load at which the onset of buckling occurs.

Backward calculations for the limiting root diameter using both the screw whip and buckling considerations are shown below (using simple-simple end fixity), with speed and torque inputs based on maximum output of the NEMA 17 stepper motor used. The resulting root diameter is 4.22 mm. The next size available is a 8-mm lead screw (6.15 root diameter), which yields a minimum FOS of 7 when the forward calculation is calculated.

Dual HIWIN MGN12 linear rails with MGN12H carriage blocks were the chosen option for linear ways in the y-axis. FOS calculations were completed using HIWIN provided load configurations, which computed both radial and lateral loads on the bearing blocks.

Fabrication and assembly was the next step in the design process. This took place from mid-October 2024 to the start of November 2024. Weed Warden's 1st design required a total of only 6 custom machined components that were fabricated by the University's Machine Shop, whereas everything else was either off-the-shelf or 3D printed using my Bambu Lab P1S. All black-colored components were 3D-printed in PLA. Below are some of the images taken during assembly.

video_2024-11-05_15-14-51.avi

video_2024-11-17_17-47-28.avi

The following image shows the final build of the ME 481 prototype, during initial power-up. A temporary camera mount assembly is featured on the left leg of the robot, with its final design discussed further in the next section.

ME 482 Design Work

The prototype of Weed Warden built in ME 481 was a really good foundation, but as a team, we all agreed that it was drifting away from its intent as a prototype for a future product. We pinned down these feelings to the overall footprint of the first prototype, notably that it was significantly taller than most other automated lawncare products (i.e robotic mowers, etc).

A major driving factor for the increased height of the initial prototype was the spade bit and collection bin. The coring bit would gather a 100 mm long soil core containing the weed that would need to be ejected. The easiest solution we came up for ejection was a spade bit located over a collection bin, such that when the coring bit was lowered over the spade bit, it would break-up and eject the soil core into the bin. The additional height came from the need for the coring bit to travel overtop the equally long spade bit.

Given this, we agreed that a large part of the mechanical design work for ME 482 would focus on height reduction of the overall robot & installation of the locomotion system.

Through detailed analysis of the existing design, height reduction was to be implemented with the following changes:

Pivoted away from coring bit for now, remove spade bit. Focus on developing robot's ability to locate and position itself over a weed

Switch from 300 mm z-axis linear actuator -> 200 mm z-axis linear actuator
Switch to weed spinner bit (like similar products on Amazon) due to time constraints
Cut frame accordingly

Final CAD Layout

The following section presents a detailed overview of the final design layout of Weed Warden as a functional prototype.

The image below depicts an assembly drawing of the final CAD layout for Weed Warden. This is the second iteration of the mechanical design of the robot, with the initial design and fabrication constructed during ME 481. The design features a 4040 mm Aluminum Extrusion frame, with all purple components representing 3D-printed components. The design features a total of only five (5) custom machined aluminum components to minimize fabrication costs for the prototype.

Working Principle(s):

Lateral motion of the drill end effector is taken care of the y-axis assembly
Vertical motion of the drill end effector is taken care of the z-axis assembly
X-axis motion is taken care of the locomotion system (forward & backward motion)

Weed Warden boasts the following dimensions, making it a relatively compact device comparable in size to existing autonomous lawncare products such as autonomous lawnmowers:

Length: 550 [mm]
Width: 535 [mm]
Height: 429 [mm] (incl. 50 mm ground clearance)

Frame Assembly

A total of eleven (11) pieces of 40x40 [mm] aluminum extrusion make up the frame of Weed Warden, with an additional 4 pieces used in the y-axis assembly. All extrusions are joined together through either 3-way corner end brackets (not pictured) or internal corner brackets (not pictured) to minimize bracket interference with usable space along the frame. The choice was made to utilize aluminum extrusions, especially at this prototype stage, due to modularity (easy to reconfigure) and ease of assembly.

Y-Axis Assembly

The y-axis assembly was designed as a fully separate module, complete with its own aluminum extrusion frame, to allow for ease of repair, modification and assembly. It is responsible for the actuation of the z-axis assembly (and subsequent drill assembly) laterally across the width of the robot. It allows for up to 200 mm of travel.

Y-Axis Linear Motion System Details:

Dual HIWIN MGN12 linear rails with MGN12H carriage blocks for enhanced precision and compact design (as opposed to linear motion shafts)
Aluminum mounting plate (WW0006), used to connect the z-axis subassembly to the y-axis linear rail system
Tr8x8 lead screw, radially supported by two 3D-printed bearing blocks to achieve simple-simple end fixity
NEMA 17 motor used to power the y-axis drive through rotation of the Tr8x8 lead screw
T8 ACME nut is used to convert rotational motion of the lead screw to lateral movement of the y-axis carriage along the rails

Other Design Details:

End stops on either end of the y-axis assembly (Component #9 & #10) double as limit switch holders used during system start-up and homing
3D-printed cable chains for effective wire management during constant operation
3D-printed guards to remove pinch hazards and house limit switches

Z-Axis Assembly

To reduce mechanical complexity and R&D time, the choice was made to use an off-the-shelf 200 [mm] linear stage actuator equipped with a NEMA 23 stepper motor and ball screw linear for the actuation of the drill assembly in the vertical axis. The z-axis assembly also features limit switch end stops on the left hand side (as shown) for homing algorithms and a cable chain for cable protection/management.

Camera Frame Assembly

The camera frame assembly's primary function is to position the Pi Camera Module 3 (CM3) at a optimal height to enable an FOV that provides sight of incoming objects as well as the position of the drill end-effector. The frame assembly features a 3D-printed construction from PLA, with the use of rubber M2 rubber standoffs (found in FPV drones) to minimize impact of vibrations on the images collected from the CM3. The assembly also features a holder for the NeoPixel LED ring light for consistent illumination of the target area for the CV model. Due to the limited space, complex 3D-printed geometry was required to affix it to the front motor mounts. The annotated images below show detailed views of the final camera frame assembly.

Locomotion System

The complete locomotion system is a 3-wheeled, FWD robot platform. I was not primarily responsible for the design of the locomotion system, so I have highlighted its final configuration below.

Front Locomotion: Given the FWD setup of Weed Warden, the front locomotion assembly required the greatest design work. The design consisted of a 3D-printed motor mount, printed in ABS due to its greater impact strength and heat deflection temperature. A straight-cut bevel gear system was used to allow parallel mounting of the Brushless DC motor relative to the aluminum extrusion. The shaft was bolted to a PLA rim using an aluminum flange, with the tire being cast from 30A silicon. To cast the silicon, 3D-printed molds were made. The choice was made to go with Silicon tires due to the increased slippage we experienced with TPU-printed tires, which were originally used due to its ease of manufacturing.

Rear Locomotion: The rear locomotion assembly was relatively simple. It featured a single, low-profile caster wheel from McMaster-Carr, chosen due to its small swivel offset, allowing the wheel to be mounted as close as possible to the frame. A 3-wheeled, rear caster setup was the preferred layout due to its minimal kinematic constraints.

Front Locomotion Assembly

Rear Caster Wheel Assembly

The following images depict the re-assembly of Weed Warden with all relevant modifications and electrical integration. A majority of this assembly was completed ahead of the February 2025 reading week, to allow Logan to fully commit to autonomy bring-up for the remainder of the term. Significant mechanical improvements, especially around the reliability of the locomotion system were made during this time as well.

Testing & Validation

Validation of the final build of Weed Warden was completed within the last 2 weeks before the symposium day in March 2025. A master test plan was authored by Ethan & Eric, and was completed by all 4 of us. A total of 17 tests were completed to ensure the final design achieved all specifications outlined in the EDS. A sample test plan is shown below, along with the verification matrix summarizing all testing results. Further details provided in the ME 482 report provided below.

Symposium Day

The 2025 Mechanical Engineering Symposium at UW was held on March 24, 2025. This is the day we got to showcase our final prototype of Weed Warden to our peers, teaching staff and judges. We had a piece of turf with fake dandelions on it to demonstrate the robot's weed locating and positioning abilities, with the robot running a continuous circuit at the push of a button.

Overall, feedback we received was extremely positive, with many saying it was one of the best projects they had seen on the day. Additionally, we received the Best Prototype Award with a $500 cash prize.

Design Reports

The following design reports describe the entire engineering process undertaken during both ME 481 and ME 482, respectively, to complete the development of Weed Warden's functional prototype.

Group 19 - Final Report - 481.pdf

Group 19 - Final Report - 482.pdf

Page updated

Google Sites

Report abuse