Recent Submissions

  • QuPWM: Quantization-Based Position Weight Matrix Feature Extraction Method for Epileptic Spikes Detection

    Bin Jardan, Joori (2022-03-01) [Poster]
    Globally, epilepsy is one of the major neurological disorders. According to World Health Organization, it is estimated that approximately 70% of people who suffer epilepsy could live seizure-free life if it was diagnosed and treated properly. After an epileptic seizure, an abnormal neuronal discharge that originates from one or more cranial lobes happens in the brain known as inter-ictal spiking. Often, this abnormal activity interferes with the affected lobe s normal activity while traveling among different lobes heralding the onset of electrographic seizure. Nowadays, the detection of such irregular neural activity is done using electroencephalography (EEG) and magnetoencephalography (MEG) manual records reading, which is time-consuming and highly prone to errors causing delayed and inappropriate intervention. Therefore, this project aims to integrate the current detection method with an improved machine learning approach that uses Quantization and Position Weight Matrix (QuPWM) to provide an autonomous, time-effective method of epileptic spikes classification from both EEG and MEG signals. Our method, QuPWM, generates useful clinical features from the framed raw signals by using the position weight matrix combined with uniform quantization, after converting them into frequency domain through a Fast Fourier Transform (FFT). Then, feeding these clinical features to the Logistic Regression (LR) classifier for inter-ictal spikes detection. This evaluation was able to achieve a maximum accuracy of 96.09%, using the motif-based PWM (mPWM) features method and the 5-folds cross-validation (CV) technique that is applied on 4 patients from the dataset with a sliding frame of size 95 sample- points and a step size of 8 sample-points. In conclusion, we were able to incorporate machine learning with EEG and MEG recordings to facilitate and provide more accurate diagnoses and treatment of epilepsy, enhancing epileptic patients life thereby.
  • The semantic PHD filter for multi-class target tracking

    Chen, Jun (2022-03-01) [Poster]
    In order for a mobile robot to be able to effectively operate in complex, dynamic environments it must be capable of understanding both where and what the objects around them are. In this paper we introduce the semantic probability hypothesis density (SPHD) filter, which allows robots to simultaneously track multiple classes of targets despite measurement uncertainty, including false positive detections, false negative detections, measurement noise, and target misclassification. The SPHD filter is capable of incorporating a different motion model for each type of target and of functioning in situations where the number of targets is unknown and time-varying. To demonstrate the efficacy of the SPHD filter, we conduct both simulated and hardware tests with multiple target types containing both static and dynamic targets. We show that the SPHD filter allows effective tracking of multiple classes of targets even with detection error to some level, and performs better than a collection of PHD filters running in parallel, one for each target class. We also provide a detailed methodology that practitioners can use to fit the probabilistic sensor models necessary to run the SPHD filter.
  • Neural Network-based Carotid-to-Femoral Pulse Wave Velocity Estimation Using PPG Signal

    Bin Jardan, Khouzama (2022-03-01) [Poster]
    The number of annual mortalities have been significantly increasing, with cardiovascular diseases (CVDs) accounting for third of them globally and becoming the leading cause of this phenomenon. Early Diagnosis of CVDs offers early intervention and consequently death prevention. Therefore, developing fast, accurate, and easily accessible diagnostic techniques is essential. For determining CVDs onset and development, several cardiovascular clinical features are utilized. The increment in arterial stiffness is a key clinical feature and a primary factor in predicting CVDs . A gold standard measurement for this clinical feature is evaluating the carotid-to-femoral pulse wave velocity (cf-PWV); however, its clinical assessment is very intrusive and complicated. One method of non-invasively, economical, and low-power-consuming monitoring the arterial stiffness is a straightforward optical technique called photoplethysmogram (PPG). In this project, PPG was incorporated with an artificial intelligence-based tool to extract central arterial stiffness by estimating cf-PWV to predict CVDs. Starting with a collected single PPG waveform at three different measurement sites: radial, digital, and brachial arteries. In addition to selected features, that were calculated using fiducial points from the PPG signal along with its first, second, and third derivatives, after feature selection and identification methods were applied. The prediction s performance is assessed by the R2 (correlation coefficient) and MAPE (mean absolute percentage error) values, reflecting good results when it is close to 1 and 0, respectively. The results, using supervised deep learning model, and numerous iterations of the database s training set, demonstrate good estimation performances utilizing the extracted features from PPG signal at the level of radial, digital, and brachial arteries with an R2 around 0.98, 0.97 and 0.95, and MAPE less than 1.71%, 1.88% and 2.22%, for each distal level respectively. Thereby, an innovative machine learning tool was incorporated to a non-invasive, accessible, and economical diagnostic technique making it user-friendly to predict CVDs.
  • RFID Sensing Method Using Signal Crack Capacitor For Untethered Soft Robotic

    Nesser, Hussein (2022-03-01) [Poster]
    Due to their softness, soft robots are inherently safe and adapt well to unstructured environments. However, they can be prone to various damage types. Therefore, Soft robots require soft, stretchable, and conformable sensors to preserve their adaptivity and safety. Soft sensors can help a soft robot perceive and act upon its immediate environment. The concept of integrating sensing capabilities into soft robotic systems is becoming increasingly important. One challenge is that most of the existing soft sensors have a requirement to be hardwired to power supplies or external data processing equipment. This work aims to remove such wires and hard electronics related to strain sensing, enabling the robot to move more freely and adding a chance to get totally untethered soft robotics. We present a wireless sensor that can be embedded, for example, in soft pneumatic actuators and observe very low strain (<0.1%) with high accuracy. This study fabricates and characterizes a new soft sensor based on chipless radio-frequency identification (RFID) tag technologies. We designed this sensor with a flexible capacitor connected to a coil to form an LC oscillator. An external coil can briefly power the LC device with electromagnetic waves to take sensor readings. This chipless RFID system can detect strain by following resonant frequency changes in response to an applied strain [1]. The capacitance part of the sensor is based on a signal crack created in a conductive film that forms a capacitor when the crack opens under strain. The advantage of this signal crack capacitance is its sensitivity, thanks to the crack's narrow opening, which significantly affects the capacitance change even for low strain (<0.1%). This signal crack capacitance has shown great sensitivity, especially for low strain detection, represented by its important Gauge factor (i.e., 63 at 0.1% strain). Adding a planer coil to the single crack capacitor, we obtain a passive and wireless device, and as such, it does not require a power supply and is capable of transporting data without a wired connection. This strain sensor is best understood as an RFID tag antenna; it shows a resonant frequency change from approximately 276 to 322?MHz upon an applied strain change from 0% to 1%. This work is the first step towards monitoring and obtaining movement information from the soft robot without rigid and expensive equipment. The dense distribution of these strain sensors can be a potential technological solution to obtain the required high-dimensional sensory information unobtrusively.
  • Constructing the Future: Intelligent, Building-Scale Additive Manufacturing

    Parrott, Brian (2022-03-01) [Poster]
    The building and construction sector is far less automated and mechanized than most industries. While the unstructured environment and lack of assembly-line repeatability provide some justification for the sector s historical reliance on skilled laborers, advancements in robotics, especially in additive manufacturing technologies, are enabling breakthroughs that may enable high-quality, low-cost and customizable construction using intelligent, building-scale additive manufacturing systems. However, current challenges related to materials performance, reinforcement requirements, and actual printing techniques have so far limited applications of large scale additive manufacturing to primarily experiments and demonstrations that do not yet deliver on the potential of this emerging technology. A new concept is presented involving an actuated print head that could significantly enhance the feasibility of large-scale additive manufacturing applications.
  • Control Design for Reduced-G Atmospheric Flights

    Chen, Yi-Hsuan (2022-03-01) [Poster]
    This work presents a longitudinal flight control algorithm to direct the aircraft to achieve microgravity through parabolic flight. The proposed control strategy makes an aircraft follow the desired trajectory by eliminating the local position errors between aircraft and proof mass. As an aircraft flies along the parabolic trajectories, it will be in a free-fall stage, thus causing the sensation of weightlessness. Moreover, triple-integral control is developed to reject the unknown, quadratically increasing aerodynamic drag. The position error is re-defined to avoid the non-minimum phase dynamics, and a PID-based controller is applied to reduce the normal position error. Simulations validate that the proposed control framework can eliminate position errors as well as reject unknown nonlinear aerodynamic drag.
  • Experiments in Robotic Self-Repair

    Caballero, Renzo (2022-03-01) [Poster]
    We observe an experiment for self-repair in robots that can fabricate their own parts. A challenge emerges when imperfections or degradations in the robot impact its ability to fabricate ideal components to guarantee self-repair. In the proposed experiment, we start with a defective or degraded component. We do not fabricate an ideal part initially, but only after a sequence of increasing-in-quality parts. We construct and validate two mathematical models to describe the experiment and match the observations. We observe convergence to the ideal part in all experiments and models, thus restoring the capability for self-repair to the robot.
  • Tetracopter: a modular fractal drone

    Wali, Obadah (2022-03-01) [Poster]
    A Tetracopter is a fractal modular drone. A characteristic of fractals is to have a geometry that can be assembled to generate the same geometry with larger scale. To achieve this feature in drones a tetrahedron geometry has been used, where a tetrahedron shape is a three side facets pyramid. The advantage of this drone structure is to have modular drones that have an assembly in vertical direction, which will increase the rigidity of the structure, as well as to reduce the wake interaction of the elevated propellers in the assembly. This work includes a design and analysis of the Tetracopter, followed by a flight testing for one Tetracopter, as well as four assembled Tetracopters.
  • Walking Robot with decoupled motion. Optimal design and synthesis

    Ibrayev, Sayat; Ibrayeva, Arman; Ibrayev, AibarIbrayeva (2022-03-01) [Poster]
    Legged machines that can adapt to off-road conditions have a number of advantages over traditional wheeled and tracked vehicles. In particular, they cause the least damage to the soil, in comparison with other vehicles, due to the discrete track on the ground, what make them beneficial in agricultural application. Meanwhile, attaining to increase productivity of agricultural machines leads to increase of machine weight that cause increased soil pressure and unfavorable conditions for plant growth. However, many of traditional (existing) designs of walking robots (WR) are extremely ineffective in terms of power consumption and complexity of control system. In particular, more than ten actuators and multi-level control are used for turning such systems. The alternative design is to consider WR as consisting of functional blocks as a) the locomotor system; b) the adaptation mechanism; c) and the mechanism of turning/maneuring - and decouple the actuators so that each actuator has a specified functional destination/purpose In this work, a rational structure of WR is proposed that allows to carry out the robot hull shifting and rotation with a minimum number of actuators. At the same time adjusting the chassis height and the foot adaptation on the rough terrain is carried out by another group of actuators. A kinematically equivalent scheme of the robot is proposed in order to simplify the study of the turning modes that allowed to determine the optimal geometrical parameters of the robot. Multicriteria synthesis methods of closed-loop robot manipulators based on isotropy, maneurability and other criteria were developed for structural-parametric synthesis of turning and develop an optimal leg mechanism, limb adaptation system on surface irregularities. The design proposed allowed to solve the problem of redundant constraints of existing designs and eliminate spurious/parasitic loads on actuators associated with multiple static indeterminacy; to eliminate additional power consumption for foot slippage; and reduce reactions in the leg joints while turning. A layout of LR was developed for experimental studies.
  • Advanced SIRD model for predicting epidemic propagation

    Ashrafyan, Yuri; Bakaryan, Tigran; Alghamdi, Abdullah; Gomes, Diogo (2022-03-01) [Poster]
    The sudden outbreak of the coronavirus disease (COVID-19) has created a public health crisis and impacted the world economy. We could contain the virus and have a low virus death rate if the cases were less than the capacity of hospitals. Hence, the forecast of the virus evolution iscrucial. However, tackling the spread of this virus is very challenging. This has increased the interest in mathematical models that help health officials and governments implement measures that mitigate the spread of COVID-19. One well-known model for predicting infectious diseases is the SIRD model. Although this model is powerful, it is not entirely accurate because it does not consider many factors such as travel, vaccination, and birth rates, which affect the total number of cases. Here, we present an improved version of the SIRD (Susceptible, Infectious, Recovered, or Deceased) model. This model is a non-linear discrete dynamical system that incorporates multiple parameters to make the prediction more accurate. Furthermore, we developed an interactive code that illustrates how the model behaves with changes in the parameters. Representative plots are included in the study to prove the adaptability and effectiveness of the model. Our model provides a realistic, qualitative representation of the epidemic that may help governments and health officials take action to stop the spread of the virus.
  • Meta-Active-Learning For Imbalanced Data Classification

    Fayoumi, Abdulrahman; Han, Wenkai; Gao, Xin (2022-03-01) [Poster]
    Classification algorithms are known to perform poorly when training with imbalanced datasets. We propose a learning approach that utilizes active learning and meta-learning to overcome the class imbalance problem. While the standard active learning approach for classification with imbalanced data works by iteratively (a) undersampling through selecting the most informative samples of the majority class or classes and (b) training with a more balanced dataset, our meta-active-learning approach does this while also accounting for previous experiences at each iteration. This addition helps in minimizing an effect similar to catastrophic forgetting, where the parameters of the model are biased towards patterns learned in the latest iterations. We demonstrate that our MAL approach produces state-of-the-art results by testing it on an artificially long-tail-imbalanced CIFAR10 dataset then comparing it to the active learning approach, which has already been proven to produce state-of-the-art results in previous studies.Practical deployments of our MAL learning method in binary and multi-label classification tasks theoretically include any domain in which class imbalance occurs, including robot vision, sentiment analysis, and medical diagnosis.

View more