AgBot Navigation in GNSS-Denied Canopies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current agricultural robotic systems face challenges in navigating and sampling crops under canopies where Global Navigation Satellite System (GNSS) signals are unreliable or non-existent, due to the dense and complex environment of crops, which limits their precision and effectiveness.
Innovation Solution
The Purdue AgBot system employs a combination of visual-inertial odometry (VIO) data from a tracking camera, LiDAR sensors, and a Monte Carlo Localization algorithm, along with an Extended Kalman Filter, to navigate and map the terrain, and includes a robotic arm with a gripper for physical sampling, using RGB-D cameras for vision-based guidance and deep learning algorithms for leaf detection and segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If GNSS-based navigation is used for autonomous agricultural robots, then navigation accuracy is improved in open environments, but the system becomes unreliable in dense crop canopies where GNSS signals are blocked
Solution Approach 1:
The patent introduces LiDAR and visual-inertial odometry systems as intermediary technologies that enable autonomous navigation in GNSS-denied environments. These sensors act as mediators by providing alternative means of localization and mapping when satellite signals are unavailable, allowing the robot to operate reliably in dense crop canopies while maintaining navigation accuracy through sensor fusion algorithms
Solution Approach 2:
The system dynamically changes operational parameters by switching between GNSS-based navigation in open areas and sensor fusion-based navigation in canopy environments. The patent employs adaptive sensor fusion that adjusts the weighting and utilization of LiDAR, visual-inertial odometry, and wheel odometry data based on environmental conditions, thereby maintaining navigation reliability across varying levels of environmental adaptability
2Productivity
If manual crop monitoring and sampling methods are used, then system complexity is reduced, but labor intensity and time consumption increase significantly
Solution Approach 1:
The autonomous robot performs crop monitoring and sampling operations independently without human intervention. The system autonomously navigates through fields, identifies crop rows using LiDAR and vision systems, monitors crop parameters, and executes physical sampling tasks. This self-service capability dramatically improves productivity by eliminating manual labor while the modular architecture manages system complexity through standardized components and interfaces
Solution Approach 2:
The patent replaces manual mechanical operations with automated robotic systems equipped with LiDAR sensors, cameras, and robotic manipulators. The complex tasks of navigation, crop identification, and physical sampling are substituted by integrated sensor-actuator systems that perceive the environment and execute precise mechanical actions, thereby improving productivity despite increased device complexity through the use of advanced sensing and control technologies
3Measurement precision
If traditional crop scouting methods are used, then measurement precision is adequate for general assessment, but time consumption and labor costs increase
Solution Approach 1:
The autonomous robot enables continuous crop monitoring by systematically traversing fields and collecting data without interruption. The system continuously scans crop rows using LiDAR and vision sensors, maintaining steady measurement acquisition as it moves through the environment. This continuous operation achieves high measurement precision for parameters such as crop height and row spacing while minimizing time loss through efficient path planning and uninterrupted data collection
Solution Approach 2:
The robot creates detailed digital copies of the crop environment through LiDAR point clouds and visual imagery. These digital replicas preserve precise geometric and textural information about crop parameters, enabling accurate measurement and analysis without requiring repeated physical visits. The copied data can be processed offline to extract precise measurements, reducing time consumption while maintaining measurement precision
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables the AgBot to accurately navigate and monitor crop conditions, including stalk height and radius, and perform physical sampling autonomously in GNSS-denied environments, improving precision and efficiency in crop monitoring and sampling.
Implementation Method 1
The tracking camera can be configured to generate visual-inertial odometry (VIO) data while the movable body navigates the ground terrain
Implementation Method 2
The first and second LiDAR sensors can be configured to capture a first set and a second set of LiDAR data, respectively
Implementation Method 3
The controller can be configured to generate terrain navigation instructions utilizing a Monte Carlo Localization algorithm, wherein the Monte Carlo Localization algorithm can include the VIO data from the tracking camera and the first set of LiDAR data
Implementation Method 4
The Monte Carlo Localization algorithm of the controller incorporates an Extended Kalman Filter (EKF) to generate the terrain navigation instructions
Data Source
AI summary
A robotic system navigates a terrain adjacent one or more agricultural crops. The system includes a movable body operable to navigate the ground terrain, a tracking camera configured to generate visual-inertial odometry (VIO) data, a first LiDAR sensor configured to capture a first set of LiDAR data, a second LiDAR sensor configured to capture a second set of LiDAR data, and a controller. The controller is configured to generate terrain navigation instructions utilizing a Monte Carlo Localization algorithm, wherein the Monte Carlo Localization algorithm includes the VIO data from the tracking camera and the first set of LiDAR data.