Browsing by Author "Kochersberger, Kevin Bruce"

Now showing 1 - 17 of 17
  • Thumbnail Image
    3D Path Planning for Radiation Scanning of Cargo Containers
    Braun, Patrick Douglas (Virginia Tech, 2022-10-28)
    Every year, the ports of entry of the continental United States receive millions of containers from container ships for processing. These containers contain everything that the country imports, and sometimes regulated items can be hidden inside them in attempt to smuggle them illegally into the country. Some of these items may be radioactive material meant for criminal purposes and represent a threat to national security. The containers are currently being scanned for radioactivity as they leave the port, but before leaving the port, containers can sit inside the port for weeks. It can be beneficial to scan these containers before they are picked up to catch the illegal material sooner and reduce the risk of danger to those nearby. Uncrewed Aerial Systems can be useful for scanning container stacks in container fields since they can be attached with sensors and reach heights that are difficult for humans. They can also scan autonomously, requiring less over watch from people. This thesis attempts to solve the problem of autonomous search by using an initial 3D scan of the search area to input into a 3D path planning algorithm to generate a flight path that will sufficiently scan the search area while minimizing flight time. Coverage is a main area of concern, as well is computational complexity and time. In order to maintain security of the aircraft, the path must be generated on-board the aircraft, and as such use on-board, lightweight, computers. The approach taken in this thesis is by breaking the problem down into 2D layers, and then developing paths on each layer based on where the obstacles are. In order to maximize coverage, contours are generated around the obstacles. The vertices of the contours are then treated like points to visit in a Travelling Salesman Problem. To incentivize paths that run alongside the obstacles for better radiation detection, paths that do not run close to the obstacles are given a higher cost than those that do, resulting in a cost-minimizing path planning algorithm yielding paths that stay close to obstacles. The Travelling Salesman Problem algorithm then yields the most time effective path to cover the area while maintaining a distance healthy for radiation scanning from the obstacles.
  • Thumbnail Image
    Adversarial Learning based framework for Anomaly Detection in the context of Unmanned Aerial Systems
    Bhaskar, Sandhya (Virginia Tech, 2020-06-18)
    Anomaly detection aims to identify the data samples that do not conform to a known normal (regular) behavior. As the definition of an anomaly is often ambiguous, unsupervised and semi-supervised deep learning (DL) algorithms that primarily use unlabeled datasets to model normal (regular) behaviors, are popularly studied in this context. The unmanned aerial system (UAS) can use contextual anomaly detection algorithms to identify interesting objects of concern in applications like search and rescue, disaster management, public security etc. This thesis presents a novel multi-stage framework that supports detection of frames with unknown anomalies, localization of anomalies in the detected frames, and validation of detected frames for incremental semi-supervised learning, with the help of a human operator. The proposed architecture is tested on two new datasets collected for a UAV-based system. In order to detect and localize anomalies, it is important to both model the normal data distribution accurately as well as formulate powerful discriminant (anomaly scoring) techniques. We implement a generative adversarial network (GAN)-based anomaly detection architecture to study the effect of loss terms and regularization on the modeling of normal (regular) data and arrive at the most effective anomaly scoring method for the given application. Following this, we use incremental semi-supervised learning techniques that utilize a small set of labeled data (obtained through validation from a human operator), with large unlabeled datasets to improve the knowledge-base of the anomaly detection system.
  • Thumbnail Image
    Analysis of Sensing Technologies for Collision Avoidance for Small Rotary-Wing Uncrewed Aerial Vehicles
    Gandhi, Manav (Virginia Tech, 2022-06-22)
    As UASs (Uncrewed Aerial System) are further integrated into operations, the need for on-board environmental perception and sensing is necessitated. An accurate and reliable creation of a 3D map resembling an aircraft's surrounding is crucial for accurate collision avoidance and path planning. Consumer UASs are now being equipped with sensors to fulfill such a requirement – but no system has been proven as capable of being fully relied upon. With many sensing options available, there are several constraints regarding size, weight, and cost that must be considered when developing a sensing solution. Additionally, the robustness of the system must not be diminished when moving to a system that minimizes size, weight, or cost. An analysis of different sensing technologies that small rotary-wing aircraft (below 25kg) can be outfitted with for collision avoidance is performed. Several sensing technologies are initially compared through technology analyses and controlled experiments. The topmost systems were then further integrated onto a small low-cost quadcopter for flight testing and data acquisition. Ultimately, a fusion between stereo vision imagery and radar was deemed the most reliable method for providing environmental data needed for collision avoidance.
  • Thumbnail Image
    Collaborative Path Planning and Control for Ground Agents Via Photography Collected by Unmanned Aerial Vehicles
    Wood, Sami Warren (Virginia Tech, 2022-06-24)
    Natural disasters damage infrastructure and create significant obstacles to humanitarian aid efforts. Roads may become unusable, hindering or halting efforts to provide food, water, shelter, and life-saving emergency care. Finding a safe route during a disaster is especially difficult because as the disaster unfolds, the usability of roads and other infrastructure can change quickly, rendering most navigation services useless. With the proliferation of cheap cameras and unmanned aerial vehicles [UAVs], the rapid collection of aerial data after a natural disaster has become increasingly common. This data can be used to quickly appraise the damage to critical infrastructure, which can help solve navigational and logistical problems that may arise after the disaster. This work focuses on a framework in which a UAV is paired with an unmanned ground vehicle [UGV]. The UAV follows the UGV with a downward-facing camera and helps the ground vehicle navigate the flooded environment. This work makes several contributions: a simulation environment is created to allow for automated data collection in hypothetical disaster scenarios. The simulation environment uses real-world satellite and elevation data to emulate natural disasters such as floods. The environment partially simulates the dynamics of the UAV and UGV, allowing agents to ex- plore during hypothetical disasters. Several semantic image segmentation models are tested for efficacy in identifying obstacles and creating cost maps for navigation within the environ- ment, as seen by the UAV. A deep homography model incorporates temporal relations across video frames to stitch cost maps together. A weighted version of a navigation algorithm is presented to plan a path through the environment. The synthesis of these modules leads to a novel framework wherein a UAV may guide a UGV safely through a disaster area.
  • Thumbnail Image
    Continuous Monitoring of High Risk Disaster Areas by Applying Change Detection to Free Satellite Imagery
    Roush, Allison Granfield (Virginia Tech, 2024-06-11)
    Natural disasters can happen anywhere causing damage to land and infrastructure. When these disasters occur in remote areas without much human traffic, it may take a long time for someone to notice that an event has occurred and to respond to it. Response time and damages could be reduced if the area could be remotely monitored. Many satellites pass over the Earth everyday collecting valuable imagery data that is free to access. However, this data can be difficult to process and use in practical applications such as monitoring an area for changes. Existing programs that use satellite imagery to monitor areas for changes can cost a significant amount of money making it inaccessible to most people. In this paper, a software program is introduced to automatically retrieve, process, and analyze free satellite imagery data and notify the user of significant changes in their area of interest (AOI). First, a software program was developed to automatically download a package of satellite imagery data from Planet Labs that met certain requirements for AOI, date, and cloud cover. A second software program was developed to download this data from the Google Cloud Storage (GCS) space and compare a current image to the composite of previous images in order to detect a change. This program then creates a figure to display the current image, the previous image, the difference area, and a summary table of the difference metrics. This figure is saved and emailed to the user if the differences are greater than the set threshold. This program is also capable of running automatically in the background of a computer every time it is logged in. The success of the program in correctly identifying areas of change was tested in three locations using historical satellite image data. The software was successful in identifying areas of change and delivering this information to the user in an easy to understand summary figure. Overall, the software was able to utilize free satellite imagery to detect changes in disaster areas and deliver a summary report to a user to take action showing that this software could be used in the future as an easy way to monitor disaster areas.
  • Thumbnail Image
    Development of Tools for Conceptual Design of a Wildland Firefighting UAV
    Newton, Nicholas James (Virginia Tech, 2023-08-03)
    The current uses of unmanned aerial vehicles (UAVs) in wildland firefighting center around mapping, scouting, and firing operations. These operations and additional operations are often held back by lack of range and lift capacity of current UAV options. Software design tools were developed in this research to aid in designing a UAV for wildland firefighting. The tools help create a mission profile, estimate the mass of the UAV, select a motor and rotor, select a battery, and generate and analyze a finite element (FE) sector model. These tools leverage parametric analysis and studying existing hardware to create a design. The FE model is generated based on the mission profile, a motor and rotor, and battery as design parameters and a set of design variables. The tools developed for creating a mission profile, estimating mass, selecting a motor and rotor, and selecting a battery successfully aid the preliminary design of an octocopter, hexacopter, and quadcopter. The FE tool was designed around an octocopter's geometry, which leads to complications in generating FE models for a hexacopter or quadcopter. Recommendations were made for altering the FE tool to account for hexacopters and quadcopters. Other recommendations were made to support future work in creating an optimized design of a wildland firefighting UAV.
  • Thumbnail Image
    An Exploration of Rapid Tooling in Low-Cost Bead Foam Molding Applications
    Dejager, Matthew Emerson (Virginia Tech, 2024-02-07)
    Many manufacturing processes require complex tooling which contributes significantly to the cost and time required to develop new products. Bead foam molding is often hampered by these limitations. This thesis presents an analysis of Additive Manufacturing (AM) applications in low cost bead foam molding, focusing on molding trials, economic analysis, and future potential. Through molding trials, the thesis evaluates the efficacy of AM tooling in comparison to traditional aluminum tooling, specifically in evaluating tool life and cost. A key finding is a reduction in lead time up to 70% and cost of up to 63% compared to traditional tooling, particularly in low-volume production scenarios. This thesis includes a detailed cost analysis, which breaks down the cost components associated with AM processes such as pre-processing, production, material costs, post-processing, and overheads. This analysis reveals that AM tooling can offer substantial cost savings over conventional methods, making it a viable option for specific manufacturing contexts. Findings suggest that while AM tooling shows significant promise in reducing costs and accelerating production in bead foam molding, further research is required. This research should focus on exploring the scalability of AM for larger tools and investigating the application of new and emerging AM processes and materials.
  • Thumbnail Image
    An Exploration of the Acoustic Detection and Localization of Small Uncrewed Aerial Systems
    Keller, Jonathan Charles (Virginia Tech, 2022-10-06)
    With the increasing number of small Uncrewed Aerial Systems (sUAS) in the airspace, the need for robust Detect and Avoid (DAA) technologies is clear. This is especially true when considering the potential for non-cooperative aircraft with unknown intent. Many UAS use high resolution cameras to perform omnidirectional scans of their nearby airspace to localize traffic. These scans can be quite computationally expensive and often necessitate the use of costly and heavy hardware components. Ground-based solutions such as centralized, stationary towers are often expensive, difficult to proliferate, and have the disadvantage of not being onboard the aircraft and as such not always local to the airspace conflict. A feasibility exploration of acoustic detection and localization of non-cooperative aircraft using a low-cost microphone array, computationally inexpensive beamforming algorithms, and filtering techniques, is performed. The cost of the system is minimized by utilizing widely proliferated microphone hardware originally designed for short-range voice detection, as well as a small Uncrewed Aerial Systems (sUAS) from a developmental kit. Lastly, an exploration is conducted to maximize the detection range of the microphone system. A comparison of filtering techniques to try to filter sUAS self-noise is compared to alternative methods such as a ballistic sampling period where the motors of the sUAS are momentarily turned off to reduce noise. A final recommendation of a multi-sensor suite of microphones, cameras, along with other potential sensors, is determined.
  • Thumbnail Image
    Fixed-wing Classification through Visually Perceived Motion Extraction with Time Frequency Analysis
    Chaudhry, Haseeb (Virginia Tech, 2022-01-19)
    The influx of unmanned aerial systems over the last decade has increased need for airspace awareness. Monitoring solutions such as drone detection, tracking, and classification become increasingly important to maintain compliance for regulatory and security purposes, as well as for recognizing aircraft that may not be so. Vision systems offer significant size, weight, power, and cost (SWaP-C) advantages, which motivates exploration of algorithms to further aid with monitoring performance. A method to classify aircraft using vision systems to measure their motion characteristics is explored. It builds on the assumption that at least continuous visual detection or at most visual tracking of an object of interest is already accomplished. Monocular vision is in part limited by range/scale ambiguity, where range and scale information of an object projected onto the image plane of a camera using a pin- hole model is generally lost. In an indirect effort to attempt to recover scale information via identity, classification of aircraft can aid in improvement of. These measured motion characteristics can then be used to classify the perceived object based on its unique motion profile over time, using signal classification techniques. The study is not limited to just unmanned aircraft, but includes full scale aircraft in the simulated dataset used to provide a representative set of aircraft scale and motion.
  • Thumbnail Image
    Human-UAV Collaborative Search with Concurrent Flights and Re-Tasking
    Broz, Alexander Turina (Virginia Tech, 2022-08-29)
    This thesis discusses a system that allows an operator to use two unmanned aerial vehicles (UAVs) to search an area. Prior work accomplished this in separate survey and search missions, and this work combines those two missions into one. The user conducts a search by selecting an area to survey, and the first drone flies above it, providing up to date information about the area. Points of interest (POI) are then marked by the user and investigated by the second drone. This system assumes a static and known obstacle map, and segmenting the environment during the missions leaves potential for future work. Both drones are equipped with cameras that stream video for the user to observe. A custom graphical user interface (GUI) was created to allow for the drones to be controlled. In addition to marking a search area and POI, the user can pause the drone and delete or add new POI to change the mission mid-flight. Both drones are commanded remotely by a ground station (GCS), leaving only low-level control to the onboard computers. This ground station uses a nearest neighbor solution to the travelling salesman problem and a wavefront path planner to create a path for the low altitude drone. The software architecture is based on the Robot Operating System (ROS), and the GCS uses the MAVLink messaging protocol to communicate with the drones. In addition to the system design, this paper discusses UAV human interaction and how it is applied to this system.
  • Thumbnail Image
    Incident Response Enhancements using Streamlined UAV Mission Planning, Imaging, and Object Detection
    Link, Eric Matthew (Virginia Tech, 2023-06-29)
    Systems composed of simple, reliable tools are needed to facilitate adoption of Uncrewed Aerial Vehicles (UAVs) into incident response teams. Existing systems require operators to have highly skilled level of knowledge of UAV operations, including mission planning, low-level system operation, and data analysis. In this paper, a system is introduced to reduce required operator knowledge level via streamlined mission planning, in-flight object detection, and data presentation. For mission planning, two software programs are introduced that utilize geographic data to: (1) update existing missions to a constant above ground level altitude; and (2) auto-generate missions along waterways. To test system performance, a UAV platform based on the Tarot 960 was equipped with an Nvidia Jetson TX2 computing device and a FLIR GigE camera. For demonstration of on-board object detection, the You Only Look Once v8 model was trained on mock propane tanks. A Robot Operating System package was developed to manage communication between the flight controller, camera, and object detection model. Finally, software was developed to present collected data in easy to understand interactive maps containing both detected object locations and surveyed area imagery. Several flight demonstrations were conducted to validate both the performance and usability of the system. The mission planning programs accurately adjust altitude and generate missions along waterways. While in flight, the system demonstrated the capability to take images, perform object detection, and return estimated object locations with an average accuracy of 3.5 meters. The calculated object location data was successfully formatted into interactive maps, providing incident responders with a simple visualization of target locations and surrounding environment. Overall, the system presented meets the specified objectives by reducing the required operator skill level for successful deployment of UAVs into incident response scenarios.
  • Thumbnail Image
    An Invariant Extended Kalman Filter for Indirect Wind Estimation Using a Small, Fixed-Wing Uncrewed Aerial Vehicle
    Ahmed, Zakia (Virginia Tech, 2024-06-06)
    Atmospheric sensing tasks, including measuring the thermodynamic state (pressure, temperature, and humidity) and kinematic state (wind velocity) of the atmospheric boundary layer (ABL) can aid in numerical weather prediction, help scientists assess climatological and topological features over a region, and can be incorporated into flight path planning and control of small aircraft. Small uncrewed aerial vehicles (UAVs) are becoming an attractive platform for atmospheric sensing tasks as they offer increased maneuverability and are low-cost instruments when compared to traditional atmospheric sensing methods such as ground-based weather stations and weather balloons. In situ measurements using a UAV can be obtained for the thermodynamic state of the ABL using dedicated sensors that directly measure pressure, temperature, and humidity whereas the kinematic state (wind velocity) can be measured directly, using, for example, a five-hole Pitot probe or a sonic anemometer mounted on an aircraft, or indirectly. Indirect measurement methods consider the dynamics of the aircraft and use measurements from its operational sensor suite to infer wind velocity. This work is concerned with the design of the invariant extended Kalman filter (invariant EKF) for indirect wind estimation using a small, fixed-wing uncrewed aerial vehicle. Indirect wind estimation methods are classified as model-based or model-free, where the model refers to the aerodynamic force and moment model of the considered aircraft. The invariant EKF is designed for aerodynamic model-free wind estimation using a fixed-wing UAV in horizontal-plane flight and the full six degree of freedom UAV. The design of the invariant EKF relies on leveraging the symmetries of the dynamic system in the estimation scheme to obtain more accurate estimates where convergence of the filter is guaranteed on a larger set of trajectories when compared to conventional estimation techniques, such as the conventional extended Kalman filter (EKF). The invariant EKF is applied on both simulated and experimental flight data to obtain wind velocity estimates where it is successful in providing accurate wind velocity estimates and outperforms the conventional EKF. Overall, this work demonstrates the feasibility and effectiveness of implementing an invariant EKF for aerodynamic model-free indirect wind estimation using only the available measurements from the operational sensor suite of a UAV.
  • Thumbnail Image
    Methods for Radioactive Source Localization via Uncrewed Aerial Systems
    Adams, Caleb Jeremiah (Virginia Tech, 2024-03-28)
    Uncrewed aerial systems (UAS) have steadily become more prevalent in both defense and industrial applications. Nuclear detection and deterrence is one such field that has given rise to many new opportunities for UAS operations. There is a need to research and develop methods to integrate existing radiation detection technology with UAS capable of flying low-altitude missions. This low-altitude scanning can be achieved by combining small and lightweight radiation detectors and state-of-the-art aircraft and avionics. High resolution mapping can then be conducted using the results of these scans. Significant work has been conducted in this field by both private industry and academic institutions, including the Uncrewed Systems Lab (USL) at Virginia Tech. This work seeks to expand this body of knowledge and provide practical experimental information to showcase and validate the efficacy of radiation detection via UAS. Multiple missions were conducted using samples of 137Cs and 60Co as a radioactive source. Various filtering methods were applied to the results of these missions to produce visual maps that aid in the localization of an unknown source to compare various flight parameters. In addition, significant work was conducted to characterize two radiation detectors available to the USL to provide metrics to assist in the UAS design and flight planning. Finally, the detectors were taken to Savannah River National Laboratories to conduct experiments to provide information to aid future designs and missions that wish to detect a wider variety of radioactive sources.
  • Thumbnail Image
    Modeling and Control of Dual-Motored Tail-Sitting Flying Wing Using a Fuzzy Logic Pid Controller
    Sebolt, Avery Jackson (Virginia Tech, 2022-11-08)
    With large-scale implementation of drones having begun and numerous companies competing to be among the original players in the market, there lies a large potential for novel drone designs to be created and flown. These novel designs are the ones that were largely ignored in the previous century due to the physical constraints of having a crewed cockpit, but uncrewed aerial vehicles, or UAVs, have opened a floodgate of potential design spaces that may be explored which were previously impossible. The hybrid vertical take-off and landing, VTOL, UAV is one aircraft that presents a potential solution to the classic trade-off of the traditional VTOL's range and endurance limitations versus the fixed wing's required infrastructure. An aircraft known as the Flite Test Spear is used to examine fuzzy logic control and is one such hybrid VTOL that uses large control surfaces and throttle control to maneuver itself for take-offs and landings in a tail-sitting orientation before transitioning to forward, fixed-wing flight. Current flight controllers used in operation on hybrid VTOL aircraft rely on a control law state machine where given a pre-identified aircraft state, the controller enters a transitioning maneuver that takes the aircraft from a VTOL to fixed wing flight regime, or vice versa. Each flight regime is operated by a PID controller with different gains and control input realizations. A modification to this principle is first examined by using fuzzy logic PID gain modification for increased response time and reduced overshoot. Reducing overshoot is of particular interest in this case as, on an aircraft such as this, it has the potential for entering undesirable and unrecoverable states, especially during its transition. Secondly, a mixing of the two flight controllers using a fuzzy logic system was implemented to combine the two controllers' outputs and potentially smooth this transition for safer, more efficient flight. The fuzzy logic controlled mixing of the two VTOL and fixed wing controllers was not proven to provide a more desirable response within the scope of the simulation, however, performed equally as well to that of the current state machine response. The gain scheduling fuzzy systems implemented in the controller have shown to decrease overshoot of the aircraft when given commands to different states, but respond slower than their conventional counterparts. Promise in the reduction of the overshoot error and their lightweight construction leads to the conclusion that implementation on a prototype aircraft would be worthwhile for further testing.
  • Thumbnail Image
    Robot Motions that Mitigate Uncertainty
    Toubeh, Maymoonah (Virginia Tech, 2024-10-23)
    This dissertation addresses the challenge of robot decision making in the presence of uncertainty, specifically focusing on robot motion decisions in the context of deep learning-based perception uncertainty. The first part of this dissertation introduces a risk-aware framework for path planning and assignment of multiple robots and multiple demands in unknown environments. The second part introduces a risk-aware motion model for searching for a target object in an unknown environment. To illustrate practical application, consider a situation such as disaster response or search-and-rescue, where it is imperative for ground vehicles to swiftly reach critical locations. Afterward, an agent deployed at a specified location must navigate inside a building to find a target, whether it is an object or a person. In the first problem, the terrain information is only available as an aerial georeferenced image frame. Semantic segmentation of the aerial images is performed using Bayesian deep learning techniques, creating a cost map for the safe navigation ground robots. The proposed framework also accounts for risk at a further level, using conditional value at risk (CVaR), for making risk-aware assignments between the source and goal. When the robot reaches its destination, the second problem addresses the object search task using a proposed machine learning-based intelligent motion model. A comparison of various motion models, including a simple greedy baseline, indicates that the proposed model yields more risk-aware and robust results. All in all, considering uncertainty in both systems leads to demonstrably safer decisions.
  • Thumbnail Image
    Supervised and self-supervised deep learning approaches for weed identification and soybean yield prediction
    Srivastava, Dhiraj (Virginia Tech, 2023-07-28)
    This research uncovers a novel pathway in precision agriculture, emphasizing the utilization of advanced supervised and self-supervised deep learning approaches for an innovative solution to weed detection and crop yield prediction. The study focuses on key weed species: Italian ryegrass in wheat, Palmer amaranth, and common ragweed in soybean, which are troublesome weeds in the United States. One of the most innovative components of this research is the debut of a self-supervised learning approach specifically tailored for soybean yield prediction using only unlabeled RGB images. This novel strategy presents a departure from traditional yield prediction methods that consider multiple variables, thus offering a more streamlined and efficient methodology that presents a significant contribution to the field. To address the monitoring of Italian ryegrass in wheat cultivation, a bespoke Convolutional Neural Network (CNN) model was developed. It demonstrated impressive precision and recall rates of 100% and 97.5% respectively, in accurately classifying Italian ryegrass in the wheat. Among three hyperparameter tuning methods, Bayesian optimization emerges as the most efficient, delivering optimal results in just 10 iterations, contrasting with 723 and 304 iterations required for grid search and random search respectively. Further, this study examines the performance of various classification and object detection algorithms on Unmanned Aerial Systems (UAS)-acquired images at different growth stages of soybean and Palmer amaranth. Both the Vision Transformer and EfficientNetB0 models display promising test accuracies of 97.69% and 93.26% respectively. However, considering a balance between speed and accuracy, YOLOv6s emerged as the most suitable object detection model for real-time deployment, achieving an 82.6% mean average precision (mAP) at an average inference speed of 8.28 milliseconds. Furthermore, a self-supervised contrastive learning approach was introduced for automating the labeling of Palmer amaranth and soybean. This method achieved a notable 98.5% test accuracy, indicating the potential for cost-efficient data acquisition and labeling to advance precision agriculture research. A separate study was conducted to detect common ragweed in soybean crops and the prediction of soybean yield impacted by varying weed densities. The Vision Transformer and MLP-Mixer models achieve test accuracies of 97.95% and 96.92% for weed detection, with YOLOv6 outperforming YOLOv5, attaining an mAP of 81.5% at an average inference speed of 7.05 milliseconds. Self-supervised learning-based yield prediction models reach a coefficient of determination of up to 0.80 and a correlation coefficient of 0.88 between predicted and actual yield. In conclusion, this research elucidates the transformative potential of self-supervised and supervised deep learning techniques in revolutionizing weed detection and crop yield prediction practices. Its findings significantly contribute to precision agriculture, paving the way for efficient and cost-effective site-specific weed management strategies. This, in turn, promotes reduced environmental impact and enhances the economic sustainability of farming operations.
  • Thumbnail Image
    Towards Improving and Extending Traditional Robot Autonomy with Human Guided Machine Learning
    Cesar-Tondreau, Brian (Virginia Tech, 2022-10-05)
    Traditional autonomy among robotic and other artificial agents was accomplished via automated planning methods that found a viable sequence of actions, which, if executed by an agent, would result in the successful completion of the given task(s). However, many tasks that we would like robotic agents to perform involve goals that are complex, poorly-defined, or hard to specify. Furthermore, significant amounts of data or computation are required for agents to reach reasonable performance. As a result, autonomous systems still rely on human operators to play a supervisory role to ensure that robotic operations are completed quickly and successfully. The presented work aims to improve the traditional methods of robot autonomy by developing an intuitive means for(human operators to adapt/mold the behaviors and decision making of autonomous agents) autonomous agents to leverage the flexibility and expertise of human end users. Specifically, this work shows the results of three machine learning-based approaches for modifying and extending established robot navigation behaviors and skills through human demonstration. Our first project combines Imitation learning with classical navigation software to achieve long-horizon planning and navigation that follows navigation rules specified by a human user. We show that this method can adapt a robot's navigation behavior to become more like that of a human demonstrator. Moreover, for a minimal amount of demonstration data, we find that this approach outperforms recent baselines in both navigation success rate and trajectory similarity to the demonstrator. In the second project, we introduce a method of communicating complex skills over a short-horizon task. Specifically, we explore using imitation learning to teach a robot the complex skill needed to safely navigate through negative obstacles in simulation. We find that this proposed method could imitate complex navigation behaviors and generalize to novel environments in simulation with minimal demonstration. Furthermore, we find that this method compares favorably to a classical motion planning algorithm which was modified to assign traversal cost based on the terrain slope local to the robot's current pose. Finally, we demonstrate a practical implementation of the second approach in a real-world environment. We show that the proposed method results in a policy that can generalize across differently shaped obstacles and across simulation and reality. Moreover, we show that the proposed method still outperforms the classical motion planning algorithm when tasked to navigate negative obstacles in the real world.