A. James Clark School of Engineering
Permanent URI for this communityhttp://hdl.handle.net/1903/1654
The collections in this community comprise faculty research works, as well as graduate theses and dissertations.
Browse
4 results
Search Results
Item TOWARDS EFFICIENT OCEANIC ROBOT LEARNING WITH SIMULATION(2024) LIN, Xiaomin; Aloimonos, Yiannis; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)In this dissertation, I explore the intersection of machine learning, perception, and simulation-based techniques to enhance the efficiency of underwater robotics, with a focus on oceanic tasks. My research begins with marine object detection using aerial imagery. From there, I address oyster detection using Oysternet, which leverages simulated data and Generative Adversarial Networks for sim-to-real transfer, significantly improving detection accuracy. Next, I present an oyster detection system that integrates diffusion-enhanced synthetic data with the Aqua2 biomimetic hexapedal robot, enabling real-time, on-edge detection in underwater environments. With detection models deployed locally, this system facilitates autonomous exploration. To enhance this capability, I introduce an underwater navigation framework that employs imitation learning, enabling the robot to efficiently navigate over objects of interest, such as rock and oyster reefs, without relying on localization. This approach improves information gathering while ensuring obstacle avoidance. Given that oyster habitats are often in shallow waters, I incorporate a deep learning model for real/virtual image segmentation, allowing the robot to differentiate between actual objects and water surface reflections, ensuring safe navigation. I expand on broader applications of these techniques, including olive detection for yield estimation and industrial object counting for warehouse management, using simulated imagery. In the final chapters, I address unresolved challenges, such as RGB/sonar data integration, and propose directions for future research to enhance underwater robotic learning through digital simulation further. Through these studies, I demonstrate how machine learning models and digital simulations can be used synergistically to address key challenges in underwater robotic tasks. Ultimately, this work advances the capabilities of autonomous systems to monitor and preserve marine ecosystems through efficient and robust digital simulation-based learning.Item INDOOR TARGET SEARCH, DETECTION, AND INSPECTION WITH AN AUTONOMOUS DRONE(2024) Ashry, Ahmed; Paley, Derek; Aerospace Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This thesis investigates the deployment of unmanned aerial vehicles (UAVs) in indoor search and rescue (SAR) operations, focusing on enhancing autonomy through the development and integration of advanced technological solutions. The research addresses challenges related to autonomous navigation and target inspection in indoor environments. A key contribution is the development of an autonomous inspection routine that allows UAVs to navigate to and meticulously inspect targets identified by fiducial markers, replacing manual piloted inspection. To enhance the system’s target recognition, a custom-trained object detection model identifies critical markers on targets, operating in real-time on the UAV’s onboard computer. Additionally, the thesis introduces a comprehensive mission framework that manages transitions between coverage and inspection phases, experimentally validated using a quadrotor equipped with onboard sensing and computing across various scenarios. The research also explores integration and critical analysis of state-of-the-art path planning algorithms, enhancing UAV autonomy in cluttered settings. This is supported by evaluations conducted through software-in-the-loop simulations, setting the stage for future real-world applications.Item ELECTROLOCATION-BASED OBSTACLE AVOIDANCE AND AUTONOMOUS NAVIGATION IN UNDERWATER ENVIRONMENTS(2013) Dimble, Kedar Dnyaneshwar; Humbert, James S; Aerospace Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Weakly electric fish are capable of performing obstacle avoidance in dark and complex aquatic environments efficiently using a navigation technique known as \emph{electrolocation}. That is, electric fish infer relevant information about surrounding obstacles from the perturbations that these obstacles impart to their self-generated electric field. This dissertation draws inspiration from electrolocation to demonstrate unmapped reflexive obstacle avoidance in underwater environments. The perturbation signal, called the \emph{electric image}, contains the spatial information of the perturbing objects regarding their location, size, conductivity etc. Electrostatic equations elucidate the concept of electrolocation and the mechanism of obstacle detection using electric field perturbations. Spatial decomposition of an electric image using Wide-Field Integration processing extracts relative proximity information about the obstacles. The electric field source is changed to an oscillatory one and a quasistatic approach is taken. Simulations were performed in straight tunnel, cluttered corridor and an obstacle field. Experimental validation was conducted with a setup comprising a tank, a computer-controlled gantry system and an electro-sensor. Consistency between the simulations and the experiments was maintained by recreating similar environments. Simulations using both the electrostatic and the quasistatic approach demonstrate that the algorithm is capable of performing various maneuvers like tunnel centering, wall following and clutter navigation. The experimental results agree with the simulation results and validate the efficacy of the approach in performing obstacle avoidance. The presented approach is computationally lightweight and readily implementable, making underwater autonomous navigation in real-time feasible.Item Bio-inspired VLSI Systems: from Synapse to Behavior(2008-08-04) Xu, Peng; Abshire, Pamela; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)We investigate VLSI systems using biological computational principles. The elegance of biological systems throughout the structure levels provides possible solutions to many engineering challenges. Specifically, we investigate neural systems at the synaptic level and at the sensorimotor integration level, which inspire our similar implementations in silicon. For both VLSI systems, we use floating gate MOSFETs in standard CMOS processes as nonvolatile storage elements, which enable adaptation and programmability. We propose a compact silicon stochastic synapse and methods to incorporate activity-dependent dynamics, which emulate a biological stochastic synapse. We implement and demonstrate the first silicon stochastic synapse with short-term depression by modulating the influence of noise on the circuit. The circuit exhibits true randomness and similar behavior of rate normalization and information redundancy reduction as its biological counterparts. The circuit behavior also agrees well with the theory and simulation of a circuit model based on a subtractive single release model. To understand the stochastic behavior of the silicon stochastic synapse and the stochastic operation of conventional circuits due to semiconductor technology scaling, we develop the stochastic modeling of circuits and transient analysis from the numerical solution of the stochastic model. The analytical solution of steady state distribution could be obtained from first principles. Small signal stochastic models show the interaction between noise and circuit dynamics, elucidating the effect of device parameters and biases on the stochastic behavior. We investigate optic flow wide field integration based navigation inspired from the fly in simulation, theory, and VLSI design. We generalize the framework to limited view angles. We design and test an integrated motion image sensor with on-chip optic flow estimation, adaptation, and programmable spatial filtering to directly interface with actuators for autonomous navigation. This is the first reported image sensor that uses the spatial motion pattern to extract motion parameters enabled by the mismatch compensation and programmable filters. The sensor is integrated with a ground vehicle and navigation through simple tunnel environments is demonstrated. It provides light weight and low power integrated approach to autonomous navigation of micro air vehicles.