# Title Team Members TA Documents Sponsor
9 Automatic Weeding Arm
Shuyue (Lucia) Zhang
Sophie Liu
Sowjanya Akshintala
Johan Mufuta design_document4.pdf
Sowji Akshintala [akshint2]
Sophie Liu [yiqiaol3]
Shuyue (Lucia) Zhang [shuyuez2]


For generations, humans have used manual labor to curb aggressive weeds, which leech nutrients and resources from staple crops. As agricultural demands and farm sizes grew, the industry started to heavily rely on chemical herbicides and pesticides in order to ensure maximum yields. Weed control through herbicide is recently under dramatic controversy for its carcinogenic potential [1] and environmental-contamination concerns [2]. Currently, farms use about 44 gallons of herbicide per acre in order to kill unwanted weeds [3]. This practice comes with risks. Runoff from the herbicide sprays threatens the natural ecosystem, hurting not only native plant species but poisoning some animals as well. Herbicide use has also affected human lives, as research has linked an increase in cancer with the use of glyphosate, a popular weed killer used in the industry [1]. In terms of economics, chemical crop control has been slowly bleeding farmers dry. Agrochemical companies have been selling genetically modified seeds that have herbicides in them, but this only boosts their herbicide sales over time as weeds have evolved into “superweeds” which require higher and stronger doses of chemicals to kill [4]. This ballooning effect can be clearly noted in the soy industry, where, as of 2008, 92% of soy plants had become glyphosate-resistant [5], requiring the industry to begin using genetically modified crops with herbicide and liquid herbicide in tandem. This is all while agrochemical companies have quietly quintupled their prices for both genetically modified seeds and chemical herbicide within the last two decades [6]. Ethically, herbicide use must be phased out but regressing to the use of human labor is not a realistic solution. Modern agriculture needs a way to streamline the repetitive act of finding and destroying specific plants while keeping the desired crops safe and healthy. Naturally, robotics can provide an answer which is both ethical and cost-effective in the long run.

#Solution Overview#

We propose a solution of an automatic robotic weeding arm, which can identify post-emergent weeds and cut them with an attached blunted sheer. Automated weeders do exist in the industry, but they still rely on herbicide use [7]. Since there are existing agricultural robots in the market that can navigate the difficult terrain of crop fields, such as the TerraSentia [8], we are not focusing on the robotic base. Rather, we see the arm as a potential extension of a robotic base, allowing us to target the specific problem of chemical-free weed removal. Our arm focuses on the identification of various seedling species and atomization of the weeding process.
The arm is fitted with a camera that can not only detect different seedlings through neural network training but also enables real-time video monitoring from a connected computer screen. Once the arm is able to detect the unwanted plant, it can maneuver and cut the weed with it’s motorized sheer. We decided to cut instead of pulling the weeds because cutting requires less force and it is more efficient when treating plants. To accomplish this, the arm will have 4 motorized joints with 180 degrees of freedom, allowing the arm to trim weeds on either side. The flexibility of the arm allows it to attack hard to reach plants effectively. Due to the arm’s trainability, it can also be easily repurposed to perform many different agricultural functions. For example, once the arm is able to learn from various plant databases, it could easily be used to pick fruit or trim foliage just by switching out the sheer-hand attachment for other applicable tools.

#Solution Components#

#Hardware and mechanical components

- Camera: An Arducam 5MP OV5647 Raspberry Pi camera module with motorized focus is connected to Raspberry Pi series board for image detection and real-time video monitoring.
- Motor: Four MG995 servo motors (4.8-7v) with stall torque 12-13kg/cm are used at the joints. The controlled rotation is 180 degrees (90 on each dimension), providing enough flexibility for the joints.
- Skeleton and mechanical support: The skeleton and dimensions of the arm are designed on AutoCAD and will be laser cut using 0.25’’ thick acrylic panel. Appropriate screws will be purchased and installed on the arm to provide the necessary support.
- Battery: First plan: A LP-E8 Li-ion rechargeable battery will be used to power the electrical system with 7.2V and 13Wh power. The battery is planned to be taken from a Canon 550D camera to save the cost of the project. We are planning to solder the battery holder our own to hook up the electrical system. Second plan: 4 AA batteries will be connected in series to power the system in 6v. The battery holder will be purchased online.
- Raspberry Pi Board: Raspberry Pi 3B+ will control the camera module, communicate with the microcontroller and allow us to record test runs and review them at a later time.
- Ultrasound: We are going to use an HC-SR04 ultrasound module. This module can be controlled through Raspberry Pi to detect the distance of the arm tip to the ground, as part of our robot’s weeding mechanism.
- LED: We are going to use an LED light bulb (no particular restriction) to indicate the status of the arm. For example, when weeds are not detected, the LED is green. When weeds are detected, the LED is red.
- Switch: We will include a switch to turn on the arm.
- Potentiometers: In order to make our arm going to the home position, we are planning to include 4 potentiometers [10]. Plan 1 is to mount them on PCB. Plan 2 is to mount them on the microcontroller. We have found both ways feasible and common and will decide based on prices and other factors. But we are leaning towards plan 2.
- Control Circuit: A microcontroller can be used to control the 4 motors utilized in the three joints and automatic sheers. We could possibly implement our homing mechanism through microcontroller utilizing potentiometers.
- PCB: A PCB board is going to be used to host our hardware. The elements on PCB will include: power (battery), LED, a mounted RaspberryPi, microcontroller, switch.

#Software Design

Our software subsystem aims to achieve weed detection and real-time monitoring. In order to detect and classify the weeds from other crop seedlings, we plan to build a Neural Network model. We will start with a simple model and increase the complexity to achieve higher prediction accuracy. The training dataset will be mainly based on V2 Plant Seedlings Dataset [9] from Kaggle, which contains images of crop and weed seedlings at different growth stages. We will expand the dataset by adding images taken by the camera module. In addition, by connecting Raspberry Pi board to the computer, we plan to enable real-time monitoring through computer screen to better evaluate robot arm performance.

#Criterion for Success#

- Detect and classify weed from other crop seedlings with high accuracy (>80%).
- Achieve real-time monitoring by obtaining input video through the camera module.
- Motorized joints move effectively to reach the seedling to cut.
- Motorized sheers can safely and effectively snip unwanted plants.
- Achieve effective and efficient hardware and software communication.


[1] Emily Dixon, “Common weed killer glyphosate increases cancer risk by 41%, study says”, CNN, Feb 14 2019,

[2] Van Bruggen A.H.C., He M.M., Shin K., Mai V., Jeong K.C.,Finckh M.R., Morris J.G.Jr., “Environmental and health effects of the herbicide glyphosate”, Science of The Total Environment, V616-617, March 2018, pp 255-268,

[3] Pesticide Stewardship Calibration Formula

[4] Brooke Borel, “Weeds are winning the war against herbicide resistance”, Scientific American Sustainability, July 18 2018,

[5] Big Ag’s Dirty Little Secret

[6] Historic Fertilizer, Seed, and Chemical Costs and Projections for 2019

[7] “The autonomous robot weeder from Ecorobotix”, Ecorobotix,

[8] EarthSense, Inc.

[9] V2 Plant Seedlings Dataset.

[10] RO botic Arm Position Control

Low Cost Distributed Battery Management System

Logan Rosenmayer, Daksh Saraf

Low Cost Distributed Battery Management System

Featured Project

Web Board Link:

Block Diagram:

Members: Logan Rosenmayer (Rosenma2), Anthony Chemaly(chemaly2)

The goal of this project is to design a low cost BMS (Battery Management System) system that is flexible and modular. The BMS must ensure safe operation of lithium ion batteries by protecting the batteries from: Over temperature, overcharge, overdischarge, and overcurrent all at the cell level. Additionally, the should provide cell balancing to maintain overall pack capacity. Last a BMS should be track SOC(state of charge) and SOH (state of health) of the overall pack.

To meet these goals, we plan to integrate a MCU into each module that will handle measurements and report to the module below it. This allows for reconfiguration of battery’s, module replacements. Currently major companies that offer stackable BMSs don’t offer single cell modularity, require software adjustments and require sense wires to be ran back to the centralized IC. Our proposed solution will be able to remain in the same price range as other centralized solutions by utilizing mass produced general purpose microcontrollers and opto-isolators. This project carries a mix of hardware and software challenges. The software side will consist of communication protocol design, interrupt/sleep cycles, and power management. Hardware will consist of communication level shifting, MCU selection, battery voltage and current monitoring circuits, DC/DC converter all with low power draws and cost. (uAs and ~$2.50 without mounting)