Abstract
Non Technical Summary
In this project, we seek to create a closed-loop system for managing manure of grazing livestock. To accomplish this goal, we will use a autonomous vehicle platform which will work collaboratively with sensors on livestock in the field. Livestock sensors will flag where and when animal defecate and urinate within the field and send this information to the autonomous vehicle. The vehicle will then plan and execute a manure management strategy based on the location of manure in the field and sensed data describing the moisture and nutrient composition of the pasture soil. Manure management options will include moving manure to different areas of the field that have more faborable nutrient composition or are less hydrologically sensitive; tiling manure into the soil to prevent surface runoff; and removing manure from the field entirely. Through these management options, they system will precision-manage the nutrient composition of soil to optimize manure value as fertilizer and minimize environmental impacts. In order to work toward this vision, we will also conduct a number of addiitonal tests, including evaluating the animal-robot interactions within the field; leveraging novel simulation platforms to efficiently train autoonomous control approaches; and development work to improve precision and accuracy of sensors to detect soil nutrient compostion. Collectively, these investigations will contribute to our efforts to generate the resultant closed-loop system which we will then demonstrate on research and working farms.
Goals / Objectives
1. Characterize grazing animal behavior int he presence of an autonomous robot within the pasture.2. Determine mechanical specifications and procedures for an autonomous robot to perform manure management tasks.3. Verify system performance within empty pastures and assess system performance within populated pastures.
1. Characterize grazing animal behavior int he presence of an autonomous robot within the pasture.2. Determine mechanical specifications and procedures for an autonomous robot to perform manure management tasks.3. Verify system performance within empty pastures and assess system performance within populated pastures.
Project Methods
Sub Aim 1.1. Develop baseline characterization of animal movements, manure deposition, and associated soil moisture and nutrient profiles without a robot present. We will house 40 cattle and 10 horses in the SmartFarm Innovation Testbed sites at the Middleburg AREC and the Shenandoah Valley AREC. These sites are equipped with surveillance cameras that continuously monitor animal behaviors. Observations will be collected for 15-d and insight gleaned from observations will be tested for consistency over the subsequent 15-d. To account for seasonal variation, this exercise will be performed 4 times during the 1st year of the project, once per season. During each 30-d period, animals will wear a sensor strapped around the tail head.During the grazing periods, we will measure the spatiotemporal dynamics of soil moisture and nutrient content using a handheld soil moisture probe and the LaMotte soil test kit (LaMotte STH5 Combination soil outfit; LaMotte Co., Chestertown, MD to measure N, P, and organic matter from grid samples collected at 10m resolution throughout the testbed. Moisture data will be used to understand the spatial and temporal nature of hydrologically sensitive areas throughout the field. Soil nutrient data will be used to develop a spatial map of the soil N and P content. Together the soil nutrient and soil moisture data provide an indication of where nutrient hotspots occur (e.g., areas where hydrologic activity and high nutrient availability coincide).Sub Aim 1.2. Evaluate robot-livestock interactions to determine flight zones, time to habituation, effective habituation strategies, and management to minimize destructive behaviors.An existing autonomous snowplow robot will be configured with sensors for Sub Aims 1.2, 1.3, 2.1, and 2.2. The plow will be removed. The robot will be controlled remotely or follow a predefined path.We will evaluate robot-livestock interactions using 3 separate behavioral tests for animals that have not been habituated to the robot: stationary robot, moving robot, and flight distance test60. These tests will allow us to detect animal's sensitivity to the robot, effects of the robot on measures of stress and welfare (heart rate variability, rumination61) and rates of any destructive behaviors (charging, head butting, kicking, striking) or destructive precursors (including head lowering, nostril blowing)62 that might result in robot destruction or indicate reduced livestock welfare. For any agnostic behavioral event we detect, we will also determine the speed, angle of approach as it pertains to the animal, and proximity of the robot 10-s before the first agonistic behavior we can detect. Using a remote-controlled robot, we will evaluate livestock behavior towards a robot over four hours each in stationary and moving robot tests, and as a function of robot speed and proximity. We will evaluate flight distance as a function of robot speed, direction as it pertains to the animal, and proximity. Our time-series data will allow us to identify changes potentially indicative of habituation60,63, as well as parameters (speed, proximity to animals) of the robot that might increase destructive behaviors that could damage the robot or are indicative of decreased welfare. We will use the data from these behavioral tests to program livestock-robot interactions, including how close the robot can approach the livestock and how quickly, and how the angle of approach might affect those measures.Sub Aim 1.3. Evaluate animal movements, soil characteristics, and manure deposition when co-housed with an autonomous robot.The impact of co-housing a robot with the animals will be tested with the same experiential design (soil testing, animal wearable sensors, 2 grazing periods, 4x/year) described in Sub Aim 1.1; however, the robot will be programmed to drive a random path through the field for 4 hours, once a day to simulate manure management during the second 15-d period.Sub Aim 2.1. Capture of pasture physical characteristics and conditions. Spatial distributions of soil water, N, P, and organic matter contents are expected to change throughout the year and in response to animal defecation patterns. HB100 10Ghz RADAR sensors have demonstrated the ability to characterize soil moisture66.Data will be collected using the PSR1 in each pasture in a grid pattern (where feasible) using both IOSs and SMs immediately following the measurements collected in Sub Aim 1.1. The study of the nutrient flow within the pasture will inform the development of robot planning algorithms that may more effectively and efficiently mitigate poor flow conditions. The robot will also be configured to acquire field characteristic data using GPS, precision altimeter, 3.3-10 GHz (US/FCC models) three-dimensional radio-frequency based sensor, an array of linearly polarized broadband radar, 10GHz radar, visible and near-infrared multispectral sensors. Soil saturation patterns will be collected to determine areas of soil sensitivity and identify areas too saturated for robot movement.Sub Aim 2.2. Create robot navigation and task performance methods, procedures, and algorithms to achieve the elements of pasture maintenance. Because it can be very difficult and time-consuming to predict all possible situations that the robot may encounter, we will leverage recent advances in deep reinforcement learning that have shown the ability to address driving skill in unstructured environments77,78. Virtual environments replicating terrain measured in the pastures will be created. A simulated robot will navigate these environments with all sensor data collected. When the platform has a prohibited event (flips over or loses steering control and collides with an obstacle), data from before the event is tagged to identify it as indicating imminent failure. The data will be used to train a network for dynamic platform stability.Sub Aim 2.3. Develop robot control algorithms for the tilling and loading/moving manure. Simulation will be used to develop controllers for robots. Algorithmic and deep neural network approaches will be used to generate control systems resilient to operation in a pasture and real-time learning based on the expected versus sensed effects of control plans will be used to continually improve and/or adapt the actuation to the specific situation. Sub Aim 3.1. Model nutrient flows within pasture systems to identify best-practice reallocation strategies for manure. Data collected in aims 1.1 and 2.1 will be used to develop a livestock-environmental model41,42,85. This model will serve several purposes; 1) it determine the initial hydrological sensitivity map of pasture areas, based on terrain and soil characteristics, which will be used by PSR2 to learn which areas of the pasture are typically nutrient source areas, and 2) it will be used in a forecasting mode to predict soil nutrient and soil moisture levels40,86. Sub Aim 3.2. Implement physical robot consistent with measurements of Aims 1 & 2. As described in Aim 1, our existing mobile robot will be used to facilitate the majority of Aims 1 & 2 which will yield data critical for use in the design of robots. The robots will be developed using existing farm equipment as the base hardware. The process will begin with a Gazebo79 simulation for early testing, a benefit of using the robot operating system (ROS)/Gazebo framework.
Sub Aim 1.1. Develop baseline characterization of animal movements, manure deposition, and associated soil moisture and nutrient profiles without a robot present. We will house 40 cattle and 10 horses in the SmartFarm Innovation Testbed sites at the Middleburg AREC and the Shenandoah Valley AREC. These sites are equipped with surveillance cameras that continuously monitor animal behaviors. Observations will be collected for 15-d and insight gleaned from observations will be tested for consistency over the subsequent 15-d. To account for seasonal variation, this exercise will be performed 4 times during the 1st year of the project, once per season. During each 30-d period, animals will wear a sensor strapped around the tail head.During the grazing periods, we will measure the spatiotemporal dynamics of soil moisture and nutrient content using a handheld soil moisture probe and the LaMotte soil test kit (LaMotte STH5 Combination soil outfit; LaMotte Co., Chestertown, MD to measure N, P, and organic matter from grid samples collected at 10m resolution throughout the testbed. Moisture data will be used to understand the spatial and temporal nature of hydrologically sensitive areas throughout the field. Soil nutrient data will be used to develop a spatial map of the soil N and P content. Together the soil nutrient and soil moisture data provide an indication of where nutrient hotspots occur (e.g., areas where hydrologic activity and high nutrient availability coincide).Sub Aim 1.2. Evaluate robot-livestock interactions to determine flight zones, time to habituation, effective habituation strategies, and management to minimize destructive behaviors.An existing autonomous snowplow robot will be configured with sensors for Sub Aims 1.2, 1.3, 2.1, and 2.2. The plow will be removed. The robot will be controlled remotely or follow a predefined path.We will evaluate robot-livestock interactions using 3 separate behavioral tests for animals that have not been habituated to the robot: stationary robot, moving robot, and flight distance test60. These tests will allow us to detect animal's sensitivity to the robot, effects of the robot on measures of stress and welfare (heart rate variability, rumination61) and rates of any destructive behaviors (charging, head butting, kicking, striking) or destructive precursors (including head lowering, nostril blowing)62 that might result in robot destruction or indicate reduced livestock welfare. For any agnostic behavioral event we detect, we will also determine the speed, angle of approach as it pertains to the animal, and proximity of the robot 10-s before the first agonistic behavior we can detect. Using a remote-controlled robot, we will evaluate livestock behavior towards a robot over four hours each in stationary and moving robot tests, and as a function of robot speed and proximity. We will evaluate flight distance as a function of robot speed, direction as it pertains to the animal, and proximity. Our time-series data will allow us to identify changes potentially indicative of habituation60,63, as well as parameters (speed, proximity to animals) of the robot that might increase destructive behaviors that could damage the robot or are indicative of decreased welfare. We will use the data from these behavioral tests to program livestock-robot interactions, including how close the robot can approach the livestock and how quickly, and how the angle of approach might affect those measures.Sub Aim 1.3. Evaluate animal movements, soil characteristics, and manure deposition when co-housed with an autonomous robot.The impact of co-housing a robot with the animals will be tested with the same experiential design (soil testing, animal wearable sensors, 2 grazing periods, 4x/year) described in Sub Aim 1.1; however, the robot will be programmed to drive a random path through the field for 4 hours, once a day to simulate manure management during the second 15-d period.Sub Aim 2.1. Capture of pasture physical characteristics and conditions. Spatial distributions of soil water, N, P, and organic matter contents are expected to change throughout the year and in response to animal defecation patterns. HB100 10Ghz RADAR sensors have demonstrated the ability to characterize soil moisture66.Data will be collected using the PSR1 in each pasture in a grid pattern (where feasible) using both IOSs and SMs immediately following the measurements collected in Sub Aim 1.1. The study of the nutrient flow within the pasture will inform the development of robot planning algorithms that may more effectively and efficiently mitigate poor flow conditions. The robot will also be configured to acquire field characteristic data using GPS, precision altimeter, 3.3-10 GHz (US/FCC models) three-dimensional radio-frequency based sensor, an array of linearly polarized broadband radar, 10GHz radar, visible and near-infrared multispectral sensors. Soil saturation patterns will be collected to determine areas of soil sensitivity and identify areas too saturated for robot movement.Sub Aim 2.2. Create robot navigation and task performance methods, procedures, and algorithms to achieve the elements of pasture maintenance. Because it can be very difficult and time-consuming to predict all possible situations that the robot may encounter, we will leverage recent advances in deep reinforcement learning that have shown the ability to address driving skill in unstructured environments77,78. Virtual environments replicating terrain measured in the pastures will be created. A simulated robot will navigate these environments with all sensor data collected. When the platform has a prohibited event (flips over or loses steering control and collides with an obstacle), data from before the event is tagged to identify it as indicating imminent failure. The data will be used to train a network for dynamic platform stability.Sub Aim 2.3. Develop robot control algorithms for the tilling and loading/moving manure. Simulation will be used to develop controllers for robots. Algorithmic and deep neural network approaches will be used to generate control systems resilient to operation in a pasture and real-time learning based on the expected versus sensed effects of control plans will be used to continually improve and/or adapt the actuation to the specific situation. Sub Aim 3.1. Model nutrient flows within pasture systems to identify best-practice reallocation strategies for manure. Data collected in aims 1.1 and 2.1 will be used to develop a livestock-environmental model41,42,85. This model will serve several purposes; 1) it determine the initial hydrological sensitivity map of pasture areas, based on terrain and soil characteristics, which will be used by PSR2 to learn which areas of the pasture are typically nutrient source areas, and 2) it will be used in a forecasting mode to predict soil nutrient and soil moisture levels40,86. Sub Aim 3.2. Implement physical robot consistent with measurements of Aims 1 & 2. As described in Aim 1, our existing mobile robot will be used to facilitate the majority of Aims 1 & 2 which will yield data critical for use in the design of robots. The robots will be developed using existing farm equipment as the base hardware. The process will begin with a Gazebo79 simulation for early testing, a benefit of using the robot operating system (ROS)/Gazebo framework.
Performance Period: 07/01/2021 - 06/30/2025
Institution: Virginia Polytechnic Institute
Sponsor: USDA-NIFA
Award Number: 2021-67021-34769
Feedback
Feedback
If you experience a bug or would like to see an addition or change on the current page, feel free to leave us a message.