Anti-reflection structured surfaces or coatings on large aperture optics need spatial transmission and reflection uniformity characterization. This work details the design and implementation of a linear mapping system built around a gantry for foot-scale optical windows.

The rising cost of robotic platforms has created significant challenges for engineering students and academic institutions. High prices often make these platforms unaffordable for students, limiting their access to hands-on experience in robotics. Similarly, schools face financial strain in maintaining and upgrading such systems, which can hinder their ability to provide quality education in robotics and automation. This issue has become particularly pressing as robotics plays a critical role in engineering and technology curriculum. A solution to these issues is a low-cost, low-power robotic system that balances affordability with advanced functionality. A platform is needed that integrates essential features such as ROS (Robot Operating System), LiDAR, motor control, and camera vision, providing a comprehensive tool-set for learning and experimentation. Furthermore, the system should support simple functionality for beginners, but also be able to accommodate students wanting to progress to advanced robotic applications. This paper explores the design and development process of this robotic platform, highlighting its modular design, affordability, and versatility.

This paper presents a detailed overview of an autonomous All-Terrain Vehicle (ATV) that tracks and follows a person. The ATV uses computer vision to identify and track objects, and a Jetson Orin Nano GPU and YOLOv8 to enable autonomous operation. The ATV can steer left, steer right, drive straight, speed up, slow down, and stop depending on the person's location relative to the ATV. The paper provides a comprehensive architecture for the ATV's controls and operation, including the hardware and software components. The ATV's computer vision system is based on YOLOv8, which is a state-of-the-art object detection and tracking algorithm. The paper also discusses the challenges associated with developing an autonomous ATV that can track and follow a person, including the need for accurate object detection and tracking, and the importance of safety and reliability. The proposed architecture and operation of the ATV can be used in a variety of applications, including surveillance, security, transportation, and military applications.

The autonomous vehicle industry is growing rapidly due to the quick advancements in technology. As society transitions into the era of autonomous vehicles, there will be many vehicles left behind that do not possess autonomous or self-driving capabilities. Therefore, it is desirable to retrofit these existing vehicles with sensors and actuators for autonomous functions. Ideally, integration of new features should be directly integrated into the existing design. In order to do this, understanding of the electrical and mechanical design of the vehicle is required, which in turn requires access to a manufacturer's design details that may not be publicly available. An alternative solution could be developed to work alongside that vehicle's existing hardware, rather than replace it. This work focuses on a solution where the objective is to integrate a rotary shaft encoder system through multiple layers of hardware and software abstraction. The encoder hardware integration will provide the necessary data feedback that will allow the conversion of a conventional human-driven vehicle to an autonomous controlled vehicle.

Robotics has become an increasingly relevant subject for engineering students over the past decade. This subject blends together many topics, such as artificial intelligence, localization, path-planning, machine vision, and human-machine interaction. The need to manage so many different aspects, and by extension, multiple applications, all running simultaneously in the same computing system has given rise to robotics frameworks, which are software packages aimed at providing communications interfaces, standardized messages, and various debugging and simulation tools. The most popular of these robotics frameworks is ROS (Robotics Operating System). The ubiquity of ROS in research and industry motivates that university courses begin introducing ROS in curriculum. To properly teach ROS, some hardware component is required. While educational platforms exists for ROS, such as the famous “Turtlebot”, these platforms tend to be very expensive, creating an expensive cost-to-entry for any course attempting to integrate ROS into their curriculum. To resolve this issue, this paper introduces a software architecture for ROS enabling inexpensive robotics kits, such as the RSLK from Texas Instruments, which are commonly used in university courses.

The focus of this work is to control a computing-resource-constrained robotics system using embedded machine learning (also known as tiny machine learning). The objective of this effort was to develop a potential lab exercise for students in a robotics class, with the goal of giving them a better understanding of how artificial intelligence applications can be used to control a robotics system. This paper discusses the fundamentals of embedded machine learning and the convolution neural network training process used for this project. The dataset used for this project and its preparation for use with the machine learning model is also discussed. An overview of the technology and data used for this project is provided describing hardware features used to control two motors on a robotic system using speech recognition by keyword spotting. Data validation and future improvements of the project related to the speech-controlled system are also discussed.

Autonomous robots commonly rely on global positioning systems (GPS) and easily recognizable, and re-identifiable, landmarks in order to achieve localization. These elements of infrastructure are not always available in the environments where autonomous robots operate. GPS is easily denied and obstructed due to its low power design. Finding and re-observing unique landmarks in an environment can also be problematic in landscapes with limited unique features. This work builds upon a system of deploy-able electronic way-points, known as BreadCrumbs [1]. Designed to provide path planning and localization information in outdoor environments through the use of GPS, these BreadCrumbs are expanded to be capable of self-localization and the ability to provide adaptive path-loss-exponent determination based on environmental placement via a Deep Deterministic Policy Gradient algorithm for use in RSSI based range determination. Thus, enabling the BreadCrumbs to act as easily re-identifiable landmarks in a variety of environments where GPS may not be accessible and natural landmarks are sparse. We present an analysis of this system in an indoor lab and outdoor forest trail setting.

Low-cost wireless devices often rely on the log-distance path loss model for determining the distance between two devices based on radio signal strength. The log-distance path loss model relies on a properly determined path-loss exponent to produce accurate results; however, this variable can be difficult to obtain or estimate accurately, especially indoors. While many statistical models and methods can be utilized to estimate the path-loss exponent, it is often determined experimentally, which is both impractical and error-prone. This work explores the feasibility of a novel application of Deep Deterministic Policy Gradient to learn the path-loss exponent of a small limited indoor space based on the RSSI information transmitting between three wireless beacons. This will be accomplished with no prior knowledge of the structural complexities of the indoor building.

This paper discusses a novel algorithm for solving a missing data problem in the machine learning pre-processing stage. A model built to help lenders evaluate home loans based on numerous factors by learning from available user data, is adopted in this paper as an example. If one of the factors is missing for a person in the dataset, the currently used methods delete the whole entry therefore reducing the size of the dataset and affecting the machine learning model accuracy. The novel algorithm aims to avoid losing entries for missing factors by breaking the dataset into multiple subsets, building a different machine learning model for each subset, then combining the models into one machine learning model. In this manner, the model makes use of all available data and only neglects the missing values. Overall, the new algorithm improved the prediction accuracy by 5% from 93% accuracy to 98% in the home loan example.

This paper discusses research to control the steering wheel angle of an All-Terrain Vehicle (ATV) in a forest environment using machine learning techniques and a camera feed. The research uses different semantic segmentation (S.S.) techniques to analyze visual data gathered from the ATV driven in the forest. It then uses different linear regression models to find the best suited model to predict the steering wheel angle. This paper discusses the data gathering process for the visual data and steering wheel angles. It also discusses the data preparation for the S.S. models and the S.S. models used. Then it presents the data preparation for the linear regression models used after the S.S.. Finally, the training and usage of the different linear regression models, the final results, and conclusions are represented. All models were computed offline on a Jetson Nano Nvidia GPU. Our objective is to find the best model which will allow the vehicle to drive in real-time in future trials.

As a part of a larger research project to control an All-terrain Vehicle (ATV) using Machine Learning techniques, this paper discusses the implementation of a base architecture for a Honda Rancher ATV to control the vehicle using digital signals and CAN bus messages. Previous research implemented a CAN network that connects a central processing unit to a throttle controller, a steering module, and a braking system. The previous research successfully accomplished controlling the ATV to move in a Figure-8 pattern backward, but it faced a problem in maintaining a constant speed on different inclined ramps and to make sharp turns going forward. This paper discusses the solutions to control the speed and the handlebar angle. The research includes a closed loop control system to control the ATV speed by getting current speed feedback from the ATV engine velocity sensor. The paper discusses different methodologies to determine the speed including an encoder and Global Positioning System. This paper also describes the problem with the caster angle in the previous studies with respect to the steering angle control. A new design was implemented with an additional servo motor using pulleys and aluminum wire to steer the handlebar as an average human would steer it. Testing demonstrated stable control of the ATV in different environments with the full range motion control of the handlebar.

Many outdoor autonomous robotic applications rely on Global Navigation Satellite Systems (GNSS), or more specifically Global Position Systems (GPS), to localize the robot. However, these systems typically have difficulty in providing accurate measurements under one meter. Hence, additional sensors are fused with GNSS measurements to increase the accuracy of localization. These devices can include, laser rangefinders, vision cameras, and wireless beacons. Wireless beacons require additional setup time but are an attractive option for environments where a path may need to be determined and traversed quickly by an autonomous robot. We present a system of wireless beacons dubbed Breadcrumb. The system is designed to be deployed in outdoor environments and provide localization information to autonomous robots in the form of: Breadcrumb GPS locations, wireless ranging, and path planning data. In addition, we detail a algorithm which allows us to reduce the number of GPS modules required by the Breadcrumb network such that only every other Breadcrumb requires a module, while still maintaining the GPS location of each Breadcrumb. This algorithm could also be used in the case of a GPS module failure or spotty connection in networks where every device is equipped with a GPS.

NOTE: The first page of text has been automatically extracted and included below in lieu of an abstract Embedded Wireless Networks Laboratory Instruction Abstract Wireless sensor networks are now considered commonplace in the automation and monitoring of home, industry and environmental habitats, as well as having applications in military applications, healthcare and traffic control. Before 2006 there was a lack of educational resources pertaining to the joint fields of embedded systems and wireless network theory, especially on the undergraduate level. Based on this need for education, The University of North Carolina at Charlotte created a class to focus on the theory and application of Embedded Wireless Networks including detailed understanding of RF environmental characteristics, advancements in energy efficient network designs, and design techniques for energy efficient embedded systems. With the increased popularity of wireless capability and energy efficient embedded computing, this educational course has become essential in creating in-demand engineers. The accompanying lab portion of the class has also allowed the students to become proficient in the use of monitoring equipment including Spectrum Analyzers, Function Generator and Oscilloscopes for measuring and troubleshooting wireless activity. One of the main focuses of this paper is the teaching method used to help graduate and undergraduate students bridge the gap between the theory of RF signals, standard wireless protocols, and hands-on applications. Theoretical studies consist of RF environments including RF signal propagation, multi-path interference, and co-existence issues in the unlicensed frequency band. Standard Protocols studied include IEEE 802.15.1 (Bluetooth), 802.15.4 (LR- WPAN including ZigBee, and IEEE 802.11. Hands-on applications consist of combining function generators with antennae and signal attenuators to study frequency components of modulated signals and effects of attenuation. While specialized courses in embedded systems and wireless networks had existed separately, layering the fields to display real-life applications gives students a deeper understanding of the components. Previous Work at other Universities Previous classes offered at other universities, including UC Berkeley7 and Yale9, covered parts of the material needed to understand the decisions needed to design and implement an embedded wireless sensor network, but not in the expanded combination of the theory and hands on lab experience. A Northeastern University4 course was highly focused on labs and projects involving wireless sensor networks for use in rescue-mission scenarios. The hardware platform used in the course included the TI MSP430 Processor and Chipcon ZigBee transceiver paired with off the shelf robot kits.

NOTE: The first page of text has been automatically extracted and included below in lieu of an abstract DESIGN OF INTELLIGENT SPACECRAFT: AN INTERDISCIPLINARY ENGINEERING EDUCATION COURSE Abstract This paper discusses a highly interdisciplinary course offered to students during the Spring 2007 semester : Design of Intelligent Spacecraft. The course integrates concepts from mathematics, physics, engineering and computer science for the purpose of educating 4th year undergraduate and introductory masters-level students on the design of intelligent spacecraft. Course content is divided into two pedagogically separate parts : 1. The historical development of physical models, including mathematical models for celestial mechanics and thermodynamics. 2. Application of these models for creating intelligent spacecraft, i.e., applications of these models to pattern recognition, computer vision, and image processing. The first section introduces physical mathematical models which, in the second section of the course, are re-visited to allow for model-based design. In part (1), a new tact is taken for teaching the historical development of mathematics and physics that shapes the scientific view of the world today. Lectures seek to emphasize the rationale behind scientific thought through the variety of personalities that have defined it best characterized by the phrase : All science was new at some point. Specific classical topics include celestial mechanics and thermodynamics which are introduced using excerpts from original works of the scientists that defined and revolutionized our understandings of these fields. Some scientists considered are Aristotle, Tycho, Kepler, Newton, Euler, Bernoulli, Fourier and other scientists relevant to course topics. Where possible, original manuscripts were provided and clarified by reformulating the work in modern terminology and mathematical notation. Historical content is complemented with discussion on contemporary space missions relevant to the discussion topic. For example, historical discussions on the discoveries of Cassini or Galileo includes discussions on the recent Cassini-Huygens mission to Saturn. Further, these discussions include mission spacecraft type, its relevant design considerations and mission objectives. Discussion of mission objectives serve to highlight current boundaries of scientific knowledge and how specific space missions seek to understand topics at these boundaries. In part (2), students implemented programs relevant to spacecraft design. Programs included phys- ical simulations of celestial mechanics, thermodynamics, and signal processing programs for im- age manipulation and signal compression. Project topics reinforce topics covered in part (1) of the course. Results for physical simulations are compared against theoretically perfect results for thermodynamic simulations and established gold-standards from NASA’s HORIZONS system in

Assistive robots are used by various individuals with medical disabilities to help with tasks such as movement. A subset of these individuals are patients with the locked-in syndrome; these patients cannot communicate with a robot through traditional means, such as with a joystick. This work designs a navigation scheme which allows for an assistive robot to be controlled by patients suffering locked-in syndrome, thus allowing the patient to move about their environment. Navigation is accomplished using an algorithm that combines autonomous robot movement and communicated commands from the patient. To bridge the communication gap between the patient and robot, naturally occurring error-related potentials are used. These ERPs can be used to establish communication between the patient and robot without relying on the patient interacting with physical stimuli, such as a keyboard or joystick. The commands communicated to the robot comes in the form of a binary: correct or incorrect command in response to the movements of the robot at an intersection in a structured building. While more complicated commands can be classified from event-realted potentials (ERPs), such as directional movement, this simple command allows for fast reliable classifications and responses. To make up for the lack of complexity from patient commands, the robot is leveraged to handle tasks such as wall avoidance, while a navigation algorithm is designed to minimize the inputs required by the user when taking a commonly traveled path. The benefits of using a semi-controlled robot for navigation vs a fully autonomous robot is compared in terms of the time taken to discover and navigate an initial path to a destination. This work serves as a proof of concept for the proposed semi-autonomous navigation scheme to validate future work into the proposed design.

NOTE: The first page of text has been automatically extracted and included below in lieu of an abstract ASSESSING SENIOR DESIGN PROJECT DELIVERABLES Abstract Historically at the University of North Carolina at Charlotte (UNC Charlotte), a grade was assigned to a senior design student at the end of each of their two semesters based on one document. Therefore, students did not know how well they were progressing in the class until the end. This method of assessment also did a poor job of validating the ABET criteria for assessing the learning objectives. A new method of assessing student groups has been implemented concurrently with a new model that requires students to turn in "development documentation" throughout each semester. Each group's industry sponsor, faculty mentor, and course instructor grade these documents against a previously published rubric. This method of assessment provides plenty of feedback on the group's performance early in the semester. The original rubrics were inspired (and sometimes duplicated) from rubrics developed by another university. The faculty found that the original versions of the UNC Charlotte rubrics needed modifications due to several different reasons, mostly to encourage more design content in the documentation. This paper describes the history of this program and the development of the rubrics. Versions of the currently used rubrics are included in an appendix. 1. Introduction Capstone design courses offer engineering students an opportunity to apply the skills they have learned throughout their undergraduate education to an applied engineering project. One of the main goals of the senior design course is to engage students in a project with real world implications that are similar to those they will face once the student enters the work force. UNC Charlotte currently offers a two-semester, multi-disciplinary senior design sequence that spans all of the departments within the College of Engineering (COE). Industry-sponsored and faculty funded research efforts comprise the projects for the senior design sequence. This is particularly advantageous for the industry sponsors, since these sponsors are afforded the opportunity to initiate elective research projects in their respective areas of interest while working closely with seniors that the company may be interested in recruiting. Students prioritize their interest in available projects through analysis of posted Statements of Work and the course instructors, who represent all departments and programs in the COE, form groups with three to four students containing diverse talents that would be representative of a typical engineering team in industry.

This paper provides a technique to rate human driving skills. This technique is initiated using a combination of deep learning in computer vision and Inverse-Reinforcement Learning (IRL). The main idea is to transfer knowledge through human-machine-human. The use of machine learning aims at teaching an agent the best driving attitudes from a human expert driver, then this agent is used to teach humans once again. To achieve that goal, deep learning semantic segmentation (S.S.) efficient network (ENet) was used to detect the road. IRL was used to get the reward function from an expert driver's behavior while driving the All-Terrain Vehicle (ATV) in the woods.

NOTE: The first page of text has been automatically extracted and included below in lieu of an abstract NASA Senior Design: Mission Assurance Management Environment Abstract This paper discusses the implementation experience of our NASA senior design project. Our NASA senior design project Mission Assurance Management Environment is to increase the reliability, availability, and safety of unmanned aircraft, by focusing on implementing the Jet Propulsion Laboratory, JPL, Flight Project Practices, FPPs, and Design Practices, DPs, in an integrated software environment. This project enables the students at California State University Los Angeles to understand the function and scope of the spacecraft mission assurance activities and to make contribution to NASA ESMD. During the senior design project implementation, students work with their advisor and NASA expert to conduct the research on mission assurance management and improve their related technical background of the project, including project requirement, design, and implementation related skills. Introduction Part of the NASA vision is to initially send robots and in the end humans to explore the Moon, Mars and possibly other celestial bodies. The Exploration Systems Mission Directorate (ESMD) is dedicated to creating a constellation of new capabilities, supporting technologies and foundational research that enables sustained and affordable human and robotic exploration of the solar system. For these long duration human and robotic exploration missions that ESMD faces, the mission assurance management of the building, launching, and operating spacecraft is a crucial component. Both predictable and unpredictable risks exist in the lifetime of the spacecraft dispatched to explore our solar system. To address this challenge, we propose to improve spacecraft building and operating processes with integrated software support environments. Added components of this environment include systems engineering approaches with research focused on mission success of the flight projects and the development of all the requirements to make sure these projects minimize the unavoidable risks that such unique endeavors entail. To achieve this goal, JPL has developed a set of practices called FPPs (Flight Project Practices) and DPs (Design Practices) [1]. However, it is difficult to ensure that the work being done by the typical “one-of-a-kind” JPL projects follow the FPPs and DPs in their entirety or at least to the extent that is necessary and sufficient to minimize risks. Therefore, we investigated and integrated approaches to implement a high level software system to support JPL’s FPPs and DPs. This will effect a possibly important change in the way JPL works at minimizing the risk involved in their flight missions. Our NASA senior design project Mission Assurance Management Environment is to increase the reliability, availability, and safety of space missions, by focusing on implementing the JPL’s FPPs and DPs in an integrated software environment.