We consider the problem of selecting an optimal subset of information sources, where the goal is to identify the true state of the world from a finite set of hypotheses, based on finite observation samples from the sources. In order to characterize the learning performance, we propose a misclassification penalty framework, which aims to reduce the likelihood of acting on erroneous predictions. In a centralized Bayesian learning setting, we study the problem of selecting a minimum cost information set, while ensuring that the maximum penalty of misclassifying the true state is bounded. We prove that this combinatorial optimization problem is submodular, and establish high-probability guarantees for near-optimal performance of a standard greedy algorithm, with the associated finite sample convergence rates for the Bayesian beliefs. Next, we consider the dual problem, where one seeks to select a set of information sources under a limited budget, while trying to minimize the maximum penalty for misclassifying the true state, under finite observation samples from the sources. We prove that this problem is submodular and establish high-probability near-optimal guarantees for a standard greedy algorithm. Finally, we validate our theoretical results through some numerical simulations, and show that the greedy algorithm works well in practice.
Low-order dynamic models of turbomachinery components are developed with a new application of a previously developed multi-state graph-based modeling approach. By modeling both temperature and pressure, this approach captures first law dynamics correctly and allows for second law dynamics to also be computed. The proposed graph-based models are validated against higher fidelity models. These models will be used for an exergy-based dynamic optimization routine for parameter selection of air cycle machines.
The increased utilization of Unmanned Aerial Vehicles (UAVs) in diverse missions, from humanitarian aid to combat operations, underscores the necessity for an efficient and cost-effective development workflow for autonomous systems. Especially for defense purposes, building autonomous target recognition systems capable of detecting, identifying, and classifying adversarial agents with machine learning models requires extensive data for training. Consequently, simulation software has become an essential tool for developers seeking to assess autonomous system performance and collect data across various environments. Furthermore, the transition to real-world, application-ready systems necessitates a simulation platform that replicates not only the vehicle control algorithms but also environmental factors that affect system performance, such as lighting conditions and sensor noise.
In response to these requirements, we introduce ‘SiDG-ATRID’ (Simulator for Data Generation for Automatic Target Recognition, Identification and Detection), a simulation platform that enables the collection of high-fidelity imagery data, powered by Unreal Engine 5. The simulator supports multi-agent simulations using the AirSim API library for UAV controls and simulates commercial aircraft traffic. This framework allows for customized camera placements to record videos or photos and manage environmental conditions such as weather and lighting. Additionally, by leveraging the Cesium API for geospatial mapping, it can accurately recreate real-world environments, enhancing the realism and applicability of simulations. This integrated approach enhances the efficiency and effectiveness of synthetic data generation for detection tasks, enabling developers to easily configure simulations and collect diverse data.
The integration of technology into exercise regimens has emerged as a strategy to enhance normal human capabilities and return human motor function after injury or illness by enhancing motor learning and retention. Much research has focused on how active devices, whether confined to a lab or made into a wearable format, can apply forces at set times and conditions to optimize the process of learning. However, the focus on active force production often forces devices to either be confined to simple movements or interventions. As such, in this paper, we investigate how passive device behaviors can contribute to the process of motor learning by themselves. Our approach involves using a wearable resistance (WR) device, which is outfitted with elastic bands, to apply a force field that changes in response to a person's movements while performing exercises. We develop a method to measure the produced forces from the device without impeding the function and we characterize the device's force generation abilities. We then present a study assessing the impact of the WR device on motor learning of proper squat form compared to visual or no feedback. Biometrics such as knee and hip angles were used to monitor and assess subject performance. Our findings indicate that the force fields produced while training with the WR device can improve performance in full-body exercises similarly to a more direct visual feedback mechanism, though the improvement is not consistent across all performance metrics. Through our research, we contribute important insights into the application of passive wearable resistance technology in practical exercise settings.
When faced with an unstructured environment filled with an unknown number and size of obstacles on a chaotic terrain, it can be a challenge to determine the best method for navigating and mapping the space. This problem, known as Simultaneous Localization and Mapping (SLAM), has typically been approached using vision-based solutions, but these solutions require clear visual conditions in order to function optimally. A different approach to sensing envioronments has been explored in soft robotic systems, specifically by sensing changes in the environment through sensing changes in the robot’s configuration. Building on this idea, we introduce a method of mapping based on colliding with and deforming around obstacles using a soft, growing robot. Instead of avoiding obstacles, as is typically done to protect robots, we take advantage of the soft, growing robot’s compliance in order to navigate through, and collect information about, the environment. Through the construction and testing of a geometry-based simulation, we analyzed the behavior and effectiveness of this approach for mapping by generating random launch positions and collecting information from contacted obstacles and traversed regions. Through a myriad of randomly generated environments, we determine that: 1) the density of obstacles in an environment has minimal impact on mapping abilities and 2) at least 70% of each environment tested can be mapped by deploying 20 or fewer soft, growing robots.
In this study, perception, planning, and control are integrated for the visual navigation of an autonomous quadrotor in an uncharted global map. To estimate the quadrotor’s pose, a visual-inertial odometry strategy is employed. Then, the semi-global matching technique is applied in the stereo vision algorithm to produce an occupancy grid, upon which both motion planning and control tasks are performed. From a comprehensive literature review, the artificial potential field method emerged as the preferred choice for motion planning due to its ability to operate in unknown global maps. To address the inherent issue of local minima, a novel solution is introduced and tested. The often observed oscillations of the artificial potential field are mitigated through a combination of minimum jerk waypoint navigation and potentials, proving to be noticeably more efficient than conventional sample-based approaches like the rapidly exploring random tree. Finally, a state feedback differential flatness-based controller is proposed based on the literature. When integrated with the generated occupancy grid and the enhanced potential field in numerical simulations, the quadrotor navigated autonomously through an unknown global map, efficiently maneuvering through narrow passages without oscillations, adeptly escaping from local minima, and reaching a goal with obstacles nearby.
In this study, we propose an observer–based resilient control strategy tailored against leader-follower multi-agent systems (MASs). Our approach specifically focuses on the system’s vulnerability in the face of false-data-injection (FDI) cyberattacks, falsifying the sensors of each follower. Since the operation and functionality of MASs are strongly dependent on the inter-agent communication-based structure, MASs have an intrinsic system vulnerability against FDI cyberattacks. The potential impact of cyberattacks can be easily propagated through the entire MAS dynamics, leading to performance degradation in consensus (e.g., formation control and velocity-matching). To address this security issue, we propose an observer–based resilient control strategy that can effectively counteract the impact of FDI cyberattacks. This method utilizes the FDI attack estimator, employing the Lyapunov stability criterion and its associated linear matrix inequalities (LMIs). Finally, we present an illustrative example of a leader-follower MAS to demonstrate the effectiveness of our proposed approach.
The long-term goal of this project is to achieve safe locomotion in non-inertial environments such as ships and planes. The challenges with that prevent this goal include unfamiliar ground motion and non-linear robot dynamics. We have gathered multiple videos in real-life environments on a ship and the first objective is to analyze that data. I plan on presenting the plots I developed showing robot position on a non-inertial platform in different sea states (such as calm and aggravated sea waves). These plots were creating by analyzing a video using OpenCV, determining the right homography based on the camera position and orientation and then plotting the required x-y coordinates as if they were being viewed from a top-down view.
A novel two wheel steering self balancing robot is designed to improve locomotion ability of a traditional Segway-like self balancing robot. This robot is capable of switching between the bicycle configuration and traditional self balancing configuration. The system is nonlinear and state dependent, which makes controller design much more difficult. To tackle this problem, Lagrange's method is used to model the system and a full order dynamic model of the system is derived, a real time iterative LQR is then implemented to control the robot.
There is an increasing need to minimize electricity generation costs and carbon emission mitigation costs within energy systems, particularly in ASEAN countries where the energy demand rapidly increases, and climate change consequences could be most severe. Similar to electricity grid optimization, carbon source-to-sink mapping presents a promising approach to address this dual challenge effectively. The coupling between the electricity grid and the carbon grid occurs primarily at coal-fired power plants, which are prevalent sources of primary carbon emissions within Southeast Asia. After identifying the largest carbon storage fields in the area, this study incorporates capture, transportation, and injection costs to formulate the optimization problem to sequester carbon emissions from the top ten sources to then transport and store at the top eight fields. Concurrently, this study optimizes the regional electricity grid by considering the bus, generator, and load parameters for each country, along with the costs across the operational spectrum. Preliminary results indicate significant potential for reductions in carbon emissions and electricity costs through the proposed optimization framework. By integrating sustainability and economic considerations, this research contributes to the development of more resilient and efficient energy systems.
Soft actuators with programmable behavior has gained increasing interests for their potential to realize an ”embedded intelligence” where complex behavior can be realized with relatively simple control inputs. Harnessing this programmability is a two-fold challenge of actuator modeling and design. Our research focuses on serial Pneumatic Artificial Muscles (sPAMs), offering an improved model capturing diverse behavior across various actuator geometries. Utilizing this model, we showcase actuating a "pattern-to-pattern" shape morphing, in which we designed actuators to form an inflated beam into a desired shape. Furthermore, ongoing work explores sPAM design with elastic end constrictions, enabling non-linear responses such as bi-directional movement and snap-through transitions, expanding the range of achievable behaviors.
Eye tracking is a behavioral sensing modality used widely in human factors research. Area of Interest (AOI) matching is essential for mapping eye-gaze measures to stimuli used in experiments, but dynamic AOI matching, i.e., mapping eye-gaze measures to non-stationary stimuli, is particularly challenging to automate. The latter is becoming increasingly important for human factors research in applications such as human driving in Level 3 autonomous vehicles. Existing methods can be categorized into four different groups. First, video annotation methods rely on a human technician to define dynamic AOIs at certain keyframes, and software is used to interpolate the areas of interest in between keyframes. Second, draw-and-track methods incorporate the use of edge detectors and allow a human technician to initially define a region that the algorithm subsequently tracks. A drawback of this approach, however, is that it requires assumptions about the object’s motion that may not be valid at all times. The third group is active video methods which involve algorithms that only work with very specific and known models of the camera’s motion through space. These models presume that the camera's motion provides information that will help initialize image segmentation techniques, and if the assumptions are not met, users must initialize the region as is done in draw-and-track methods (Huang 2011). Finally, the fourth group of methods rely on image segmentation to segment images into individual objects (Panetta 202). These methods generally involve complex deep learning techniques that require the processing of individual frames and may rely on specialized hardware; thus, they are not suited for real-time applications with high-resolution displays (Lagmay 2022). Additionally, image segmentation at best can be no more precise than a pixel since it relies on the color values of individual pixels to segment the image into sets of objects. Given the limitations of these existing methods, the contribution of this work is a novel method for dynamic AOI matching that is well-suited for environments in which the locations and orientation of objects of interest in the simulated world are known. This method involves (1) defining polyhedrons in 3D simulation coordinates that enclose the object of interest in local coordinate frames, (2) transforming the vertices of these polyhedrons into the camera frame, (3) projecting these vertices into pixel space, and (4) computing the convex hull of the projected points to obtain a 2-dimensional polygon that defines the area of interest on a screen. Importantly, this approach is fully automated and does not rely on machine learning algorithms, thereby improving its speed and robustness to sources of error such as lighting that may compromise other methods. Furthermore, the speed of a direct coordinate transform approach is invariant to the size of the display, unlike image segmentation methods. To demonstrate these advantages, we consider the problem of matching gaze data to objects of interest in a driving scenario, including cars, pedestrians, traffic lights, stop signs and construction zones, that are simulated in Unreal Engine 5 and displayed on 3 1920x1080 screens for a combined resolution of 5760x1080.