Development of Novel Directed Optical Energy Weed Management Robotics Platform for Sustainable Soybean Farming

Project Overview

Project Type: On-Farm Research
Funds awarded in 2022: $20,000.00
Projected End Date: 03/31/2025
Grant Recipient: Middle Tennessee State University
Region: Southern
State: Tennessee
Principal Investigator:
hongbo zhang
Middle Tennessee State University

Information Products

Precision Optical Weed Removal Evaluation with Laser (Conference/Presentation Material)
Precision Optical Weed Recognition and Control (Conference/Presentation Material)


  • Agronomic: soybeans


  • Energy: energy use
  • Pest Management: prevention, weed ecology, robotics

    Proposal abstract:

    In the proposed research, a smart and low-cost directed optical energy weed management robotic platform for sustainable agriculture will be made to address the issue. The high throughput directed energy weed management is a novel technology for sustainable weed management. Since it does not involve the use of chemical weed removal solutions, it will enable practical weed management for producing organic food. The low-cost features of the robotics platform will enable the affordability of the robotics platform to farmers especially socially disadvantaged farmers. The smart and integrated features of the robot will ensure user-friendliness and easiness of operation and most importantly the smart decision-making capability for practical weed management.

    The novel directed energy weed management platform will be built for high throughput and diversified weed removal. In contrast to conventional weed removal methods, directed energy weed removal method does not involve the use of environmentally harmful chemical agents for sustainable weed removal. The novel robotics-based directed energy weed removal method is precise and easily automated. A new high throughput optical weed detection sensor will be built for high speed and high accuracy weed removal. A wide spectrum UV light weed removal technique will be used for improving weed removal efficiency. The directed energy weed management platform will be integrated with a low-cost robotic platform to achieve autonomous directed energy weed removal.  It is expected that the low-cost directed energy pest management platform will be adopted by small-sized farms for effective weed removal.

    The affordable autonomous weed management robotics platform is designed specifically for low-income, new, and socially disadvantaged farmers. In contrast to the conventional agricultural robot, which is often expensive to make and maintain, the proposed robot is centered by low-cost design principles for both production and maintenance thus practical to be adopted by a wide range of farmers. The extensible features of the low-cost robot will enable it conveniently interface with other equipment to construct a complete precision agriculture ecosystem. The autonomous functions of the robot from navigation to weed management will remove the barrier of technical challenges faced with new farmers thus alleviating the difficulties of new farmers for weed management.

    Directed energy is an important weed removal technique for integrated weed management. The method utilizes directed energy sources such as microwave, flaming, steaming for weed management [23, 24, 25, 26]. The method does not use harmful chemical agents and has also proven capable of weed management. While the benefits are clear, the current directed energy methods are known associated with a few fundamental challenges. First, the integration of directed energy weed management solutions with the robot is difficult.  Previous research has attempted to attach the directed energy setup to a robotics platform, but the setup is often bulky thus difficult to operate in the field. Additionally, there is little to no software integration in terms of mapping and navigation integration [27]. Second, directed energy weed management is energy inefficient. For instance, microwave weed control requires the energy of 1000 - 3400 Kg/ha rendering it an impractical weed control technology [28]. Similarly, the low thermal efficiency of flaming pest control is also a practical concern, where only 15% of energy can be effectively utilized [28]. For CO2 laser-based direct energy solution, the energy conversion efficiency is poor, about 10% of the electrical-optical conversion rate, which thus limits the actual adoption of this technology [29]. Furthermore, the inefficient energy utilization makes it difficult to strategically integrate the energy source to the robotics platform making the entire platform rather heavy in size and weight thus the short run time. Third, the speed of weed removal using a directed energy approach is slow. The current optical weed control method requires the use of either CO2 laser or an ultraviolet light bulb as the light source. For either approach, it will take a considerable amount of time from 0.5 second to 2 seconds to deliver sufficient energy for deactivation or removal of weed[29]. Fourth, the current directed energy weed control method is most effective for early stage weed removal which further limits its practical usage in practice [28, 29].

    The proposed research is built on top of the principal investigator’s previous research accomplishments. It includes research on optical sensing and laser development. It will ensure the successful conduct of the proposed directed energy optical source. The proposed robotics platform is hinged on the principal investigator’s previous experiences in building various agricultural robotic platforms thus ensuring the successful delivery of the final product. More specifically, three goals have been set up for the project.

    Goal 1: High throughput Directed Energy Source Development and Testing. Novel high throughput Directed energy equipment will be developed for effective weed management. Novel optical scanning-based wide spectrum ultraviolet light sources will be experimented with and validated. A new optical weed detection system for rapid detection of weed will be developed such that high-speed weed detection and removal will be achieved.  The efficiency of weed removal with the solution will be tested on Batey Farms.

    Goal 2: Affordable Robotics Platform Development and Testing.  The proposed platform will be at the price of 5,000$ in contrast to the average of 50,000 $ for the existing agricultural robotics platform. The affordable robot will be tested on the soybean field of Batey Farms. 

    Goal 3: Smart Robot Control, Data Analysis Pipeline Development and Testing. The importance of big data and artificial intelligence on sustainable agriculture has been well demonstrated [30, 31]. For this, a smart robot control and data analysis pipeline will be developed. Specifically, modern computer vision algorithms will be built for accurate path planning and automatic return to home for recharging. Furthermore, data analysis for the effectiveness of weed management will be presented in a user-friendly smartphone interface to control the robot and visualize the data. We will test the smart data analysis pipeline on Batey Farms asking Brandon Whitt to operate the robot using the smartphone app by himself. Problems will be identified through this process thus to improve the pipeline. 


    Project objectives from proposal:

    For achieving the three goals to maximize weed management efficiency, we propose to build the IWM robotics platform. The platform will be built based on the principal investigator’s previous research experiences [32, 33, 34, 35, 36, 37, 38]. Overall, we will utilize novel optical science, computer vision, and engineering technology for building the platform.

    1: Novel Directed Energy For Weed Control

    Non-chemical-based directed energy weed control methods will be utilized for weed control. The technology has been piloted and proven successful in the control of weed [32, 33]. The limitation and bottleneck of the technology however is a slow speed and low weed removal efficiency [39]. We propose a novel technology to enable a higher throughput weed removal system by a strong integration of new weed detection and weed removal components into one platform. Based on our recent findings, a high speed and lower cost weed detection optical system will be developed [40, 41].  Our previous research has shown that the high accuracy optical-based image classification is feasible to improve the classification accuracy to over 20%, which will serve as the base for the new high resolution based weed detection method [34, 37, 38]. To improve the weed removal efficiency further, a wide spectrum approach will be experimented for a more rapid cut or irradiation of weed for removal. This unique approach does not require increasing the power of a single light source but is able to offer stronger weed removal efficiency.

    Energy efficiency is critical for the practical application of directed energy for weed management. The wide spectrum technique will be used for increasing energy efficacy thus increasing the efficiency of thermal damages to the weed for more efficient weed removal [42]. Importantly, the energy-efficient Ultraviolet Diode will be used for weed removal to address the energy efficiency bottleneck faced with the directed energy technique. Optical scanning-based beam steering will further save energy and improve energy efficiency [43, 44, 45]. The integration of the directed energy platform into the cost-efficient robot will be also experimented to validate the robotics run time based on the energy requirement of the directed energy and the weight of the platform. The hardware interface between the directed energy platform will be also built using the CAN bus interface such that the light source can be conveniently controlled with the computer on the robotics platform.


    2: Affordable, Extensible, And Autonomous Robotics Platform

    To reach a wide range of farmers especially socially disadvantaged farmers and new farmers, the weed control robot needs to be cost-effective. To minimize the cost, the design of hardware will extensively utilize inexpensive high-strength plastic materials. The number of motors will be also minimized with optimized output torque to further save cost.   In particular,  the navigation of the robot will be relying on the PI’s recent research discovery showing the effectiveness of the single-camera approach for autonomous robot navigation [35, 36, 37]. It has been known that the cost of sensors is a significant portion of conventional robotics platforms. For example, a Velodyne Lidar sensor can cost up to 8,000$. RTK GPS costs up to 3,000$. The single-camera robot navigation approach is expected to significantly reduce the cost to the lowest. To enable practical deployment of the robot in the field, the electronics and computer hardware will be waterproofed thus reducing the reliance on weather conditions.

    The extensibility of the robotics platform is one major focus on the development of the platform. It involves the standardized hardware including the chassis, the choice of motor, the control circuit board, the standard CAN bus communication protocols with the low-cost camera, GPS, and IMU sensors. As such, it is convenient for the robot to be extended for further development and practical farm applications. With regard to software standardization, ROS software suit will be used for the operation of the robot, which will enable flexible control as well as data synchronization with other sensors.  The standardization will make the robot easier to repair and troubleshoot, making it practical for farmers to adopt it.

    3: Smart Robot Control, Data Analysis and Visualization Pipeline

    Smart data analysis is essential for effective weed management. A heavy focus of the data analysis pipeline is to further develop our existing computer vision algorithms to ensure the autonomous navigation of robots only with a single camera[35, 36, 37].  Simultaneously, the vision-based navigation method will co-work with GPS and IMU sensors for effective vision odometry to precisely guide the robot. The second major focus is to create a novel weed detection method based on our camera array Fourier Ptychography deep learning-based image classification method [38] to collect a wider field of view crop information. The collected data are further subjected to data processing to identify weed in real-time to achieve efficient weed removal. The standardization of the data analysis pipeline will be also performed. The standardized modular design of the software will be using JSON as data communication protocol. JSON data will be fed into a Lightweight Javascript-based web portal to present farmers with a friendly smartphone user interface to allow farmers to operate the robot and inform Farmer of the weed recognition results and removal efficiency. As such farmers can make timely and informative decisions on planning weed management. A smartphone app will be also built for farmers to operate the robot. Farmers will also visualize the status of the weeding process and outcome easily through the app.


    Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the view of the U.S. Department of Agriculture or SARE.