Recipient Organization
University of Utah
201 S President Circle RM 408
Salt Lake City,UT 84112-9023
Performing Department
(N/A)
Non Technical Summary
Precision management of agricultural inputs, such as fresh water, is crucial to sustainably meet increasing food production needs. However, automated systems typically require a powered infrastructure (either wired or using rechargeable batteries) to be installed and maintained which presents a barrier to widespread implementation. This project aims to flip this paradigm for precision sensing in agriculture by developing an automated soil-moisture monitoring system that consists of autonomous mobile robots, both unmanned aerial vehicles (UAVs) and quadruped ground robots, and buried passive soil sensors for monitoring soil moisture level. Importantly, the sensors require no power source, are completely underground, and can stay buried in the field for years at a time. The mobile robots will autonomously navigate to sensors, collect data to produce a soil moisture map of the field, and return to recharge themselves. In the future, this mobile robotic platform can be extended for other soil sensors (e.g., nutrients and organic carbon) and agricultural tasks such as application of herbicides and pesticides.Our primary objective is to develop and demonstrate a robotic-sensor system for precision irrigation management. This objective builds upon prior research investigating wirelessly powered soil moisture sensors. The specific objectives for this project are: 1) Design, simulate, and characterize the passive soil moisture sensing coils. 2) Develop machine-learning motion planning, navigation, and localization methods for the mobile robots to autonomously navigate to the buried sensors and localize them with sufficient precision to efficiently obtain a soil moisture measurement. 3) Develop a soil moisture mapping method utilizing multiple inputs that can provide an accurate and spatially dense map of soil water depletion.The passive soil moisture sensors make use of the relationship between soil moisture and the electrical properties, chiefly permittivity, of the soil. The proposed sensors will consist of omnidirectional coils that have a passive self-resonance frequency. This frequency changes as a function of soil moisture. The change in self-resonance frequency can be read from the above ground mobile robots (without physically contacting the sensor). Importantly, the sensors require no onboard power, are mechanically robust, and are independent of orientation. To collect soil moisture data, the mobile robots must be able to locate the buried sensors with sufficient precision. First, the robots will use onboard GPS and a map of sensor location to get relatively close to the sensor, but not close enough to collect data. Next, the final localization stage will make use of the electromagnetic coupling between a coil mounted to a 3D robotic arm on the UAV or quadruped robot and the buried sensor. This coupling signal and sensor models will be processed by a machine learning algorithm (Bayesian estimator) to home in on the location of the buried sensor. The localization phase will also exploit information-theoretic motion planning to control the motion of the robotic arm to precisely locate the sensor and collect soil moisture measurements. These measurements will then be used to produce a spatially and temporarily dense map of soil water depletion (SWD) which could be used to inform an automated irrigation management system. The sensors will directly measure volumetric water content which will then be converted to SWD. The measurements are subject to several sources of uncertainty related to sensor depth, potential sensor damage, soil type (i.e., sensors will not undergo field-specific calibrations), and slow sensor drift. To account for these sources of uncertainty, we will further exploit machine learning through Bayesian inference. First, we will quantify the approximate levels of uncertainty through basic experiments. The uncertainty values will be used to create a prior distribution, which will then be used to update the Bayesian filter. The filter will produce a "best guess" at the SWD at a particular location along with a likelihood value. The final system, including soil moisture sensors, above-ground UAVs and quadruped robots, and soil moisture mapping will be validated in field tests over one growing season.
Animal Health Component
90%
Research Effort Categories
Basic
10%
Applied
90%
Developmental
(N/A)
Goals / Objectives
The primary goal of this project is to develop and demonstrate an automated soil-moisture monitoring system for precision agriculture that consists of autonomous mobile robots (e.g., unmanned aerial vehicles (UAVs) and/or quadruped robots, referred to as 'drones'), our custom wireless powering technology, buried active and passive soil sensors, and drone-mounted temperature sensors integrated into a precision irrigation management system. With this project, we aim to flip the dominant paradigm for precision sensing in agriculture. Rather than relying on a network of powered active sensors each with their own communications hardware that form a wireless network to transmit and collect data, we will rely on mostly passive sensors and autonomous drones to collect data and relay that data to an automated management system.The three main objectives of this project are:Design, simulate, and characterize the passive soil moisture sensing coils.Develop machine-learning motion planning and localization methods for the drone to navigate to the buried sensors and localize them with sufficient precision to efficiently transfer power and/or get a soil moisture measurement.Develop a soil moisture mapping (estimation) method utilizing multiple inputs that can provide an accurate and spatially dense map of soil water depletion.
Project Methods
Our project is organized into 3 primary objectives:Design, simulate, and characterize the passive soil moisture sensing coils.Develop machine-learning motion planning and localization methods for the drone to navigate to the buried sensors and localize them with sufficient precision to efficiently transfer power and/or get a soil moisture measurement.Develop a soil moisture mapping (estimation) method utilizing multiple inputs that can provide an accurate and spatially dense map of soil water depletion.Our methods and approach for each of these three objectives are as follows.Objective 1: Sensor Design, Calibration, and CharacterizationWe will design, implement, and characterize an array of passive, wireless, and battery-free soil moisture sensors that are buried up to 15cm below the soil surface and can communicate with an above-ground drone through an inductively-coupled time-domain resonant response sensing scheme. We will characterize the calibrated output of our passive sensors in response to orientation, temperature, and moisture ingress. This characterization will be accomplished first in the lab. For all cases, we will measure a sufficient number of sensors with replicates to obtain a good estimate of part-to-part and measurement-to-measurement variability. During year 3 of the project, we will characterize the accuracy and potential drift over time of the sensors in a field study. Acclima TDR sensors will be used as a standard for comparison. Clearly in-field characterization over more than one year is desired. However, given the timeline of the present project, such longer term characterization will have to be accomplished in follow-on studies.Objective 2: Drone Navigation and Localization of SensorsWe will develop two types of mobile robot platforms, a UAV drone and a quadruped system, with the ability to autonomously navigate through a field and localize buried sensors, wirelessly charge them as needed, and collect data. Each platform has a 3D arm that carries the primary coil for fine positioning to localize the buried sensors. Given our experience and expertise with UAVs, we will first develop this type of platform because UAV-drones are currently in use in farming, for example, crop spraying, mapping, and general assessment.The second platform that will be developed and evaluated is a quadruped robot. Recent advances have led to the availability of functional quadruped robots that can potentially be used effectively to navigate through complete terrain such as a field with irrigation equipment. These systems can operate 5-10 times longer on one charge compared to traditional drones. We will acquire and use a Unitree Go1 EDU Explorer platform with 3D LiDAR sensing, wireless communication, on-board real-time computing, and an other on-board sensors such as camera. A low-cost 3D manipulator (arm) will be attached to each robot to carry and position the primary coil as means for fine positioning during the buried sensor localization process. Custom motion control, navigation and collision avoidance algorithms will be developed to enable the robots to autonomously navigate and search for buried sensors. The search process consists of two stages: (1) the robots will first use rough GPS information to get relatively close to the buried sensors. For example, the drone will land or the quadruped will stop and stand in place. (2) Using the 3D arm with attached primary coil, the fine search process will be implemented. The target estimates will then be fed into an information-theoretic motion planner to control the motion of the arm (and robot as needed) for precise localization of the buried sensor. Once localized, wireless charging and data collection will occur, and the process is repeated for other sensors in the field. The drones will report their data back to a ground station. A high-level mission planner, run on the command station, will be developed to coordinate the overall operation by optimizing which sensors in the field the drones should look for based on the available power, soil moisture data, irrigation schedule, etc.Objective 3: Soild Moisture EstimationThe overall goal of this objective is to develop and demonstrate an accurate map of soil water depletion (SWD) with sufficient spatial and temporal density to inform various types of irrigation. SWD is a measure of the amount of available water that has been removed from the soil and is a key parameter used for irrigation scheduling systems. We will use moisture sensors to provide an estimate of SWD using the measured volumetric water content. Together with the volumetric water content at field capacity (FC) and permanent wilting point (PMP), the measured volumetric water content can be used as a direct input to an irrigation scheduling system. In order to provide a spatially and temporally dense map of SWD, we will need to be robust to the following sources of uncertainty:Sensor depth.Missing or damaged sensors.Different soil types.Slow sensor drift.To account for the four sources of uncertainties noted above, we will exploit machine learning through Bayesian inference. First, we will quantify the approximate levels of the four uncertainties noted above, for example through basic experiments. The uncertainty values will be used to create a prior distribution, which will then be used to update the Bayesian filter. The filter will produce a "best guess" at the SWD at a particular location along with a likelihood value. Each time a sensor is measured, the SWD will be updated.Once passive sensors have been fully characterized in laboratory experiments, and the soil water depletion methods characterized via simulation, we will undertake a series of field tests to characterize the accuracy of the soil moisture map obtained using the methods described above. We will have access to two field sites which are irrigated by surface flood and wheel line sprinklers with Parleys loam and Kiman fine sandy loam soil, respectively. We will install a grid of 10 passive sensors in each field and one active sensor per field. These sensors will be used to provide a soil water depletion estimation map over the course of one growing season (May - September). As a ground truth measurement, a commercial Acclima TDR-310H time domain reflectometry (TDR) probe will be installed next to 5 of the 10 passive sensors. The TDR probes will undergo a field specific calibration. The accuracy of the soil moisture depletion method using the passive sensors can then be validated against the soil moisture depletion estimates using the calibrated TDR probes.