Kenneth A. McIsaac’s research while affiliated with Western University and other places

This paper presents an autonomous parking control system for an active-joint center-articulated mobile robot. We begin by proposing a kinematic model of the robot, then derive a control law designed to stabilize the vehicle's configuration within a small neighborhood of the target position. The control law is developed using Lyapunov techniques and is based on the robot's equations of motion in polar coordinates. Additionally, a beacon-based guidance system provides real-time feedback on the target's position and orientation. Simulation results demonstrate the robot's capability to start from arbitrary initial positions and orientations and successfully achieve parking.

Advances in data science and wearable robotic devices present an opportunity to improve rehabilitation outcomes. Some of these devices incorporate electromyography (EMG) electrodes that sense physiological patient activity, making it possible to develop rehabilitation systems able to assess the patient's progress when performing activities of daily living (ADLs). However, additional research is needed to improve the ability to interpret EMG signals. To address this issue, an off-line classification approach for the 26 upper-limb ADLs included in the KIN-MUS UJI dataset is presented in this paper. The ADLs were performed by 22 subjects, while seven EMG signals were recorded from their forearms. From variable-length EMG time windows, 18 features were computed, and 13 features more were extracted from frequency domain windows. The classification performance of five different machine learning techniques, including Support Vector Machines (SVM), Convolutional Neural Networks (CNN), Gated Recurrent Unit (GRU) network, XGBoost, and Random Forests, were compared. CNN performed best amongst individual models, with an accuracy above 80%, compared to SVM with 77%, GRU with 73.9%, and the tree-based models below 64%. Ensemble learning with four CNN models achieved an even higher accuracy of 86%. These results suggest that the CNN ensemble model is capable of classifying EMG signals for most ADLs, which could be used in off-line quantitative assessment of robotic rehabilitation outcomes.

Autonomous image recognition has numerous potential applications in the field of planetary science and geology. For instance, having the ability to classify images of rocks would allow geologists to have immediate feedback without having to bring back samples to the laboratory. Also, planetary rovers could classify rocks in remote places and even in other planets without needing human intervention. Shu et al. classified 9 different types of rock images using a Support Vector Machine (SVM) with the image features extracted autonomously. Through this method, the authors achieved a test accuracy of 96.71%. In this research, Convolutional Neural Networks(CNN) have been used to classify the same set of rock images. Results show that a 3-layer network obtains an average accuracy of 99.60% across 10 trials on the test set. A version of Self-taught Learning was also implemented to prove the generalizability of the features extracted by the CNN. Finally, one model has been chosen to be deployed on a mobile device to demonstrate practicality and portability. The deployed model achieves a perfect classification accuracy on the test set, while taking only 0.068 seconds to make a prediction, equivalent to about 14 frames per second.

Pixel-wise classification of hyperspectral images (HSIs) from remote sensing data is a common approach for extracting information about scenes. In recent years, approaches based on deep learning techniques have gained wide applicability. An HSI dataset can be viewed either as a collection of images, each one captured at a different wavelength, or as a collection of spectra, each one associated with a specific point (pixel). Enhanced classification accuracy is enabled if the spectral and spatial information are combined in the input vector. This allows simultaneous classification according to spectral type but also according to geometric relationships. In this study, we proposed a novel spatial feature vector which improves accuracies in pixel-wise classification. Our proposed feature vector is based on the distance transform of the pixels with respect to the dominant edges in the input HSI. In other words, we allow the location of pixels within geometric subdivisions of the dataset to modify the contribution of each pixel to the spatial feature vector. Moreover, we used the extended multi attribute profile (EMAP) features to add more geometric features to the proposed spatial feature vector. We have performed experiments with three hyperspectral datasets. In addition to the Salinas and University of Pavia datasets, which are commonly used in HSI research, we include samples from our Surrey BC dataset. Our proposed method results compares favorably to traditional algorithms as well as to some recently published deep learning-based algorithms.

Unmet expectations contribute to a high patient dissatisfaction rate following total knee replacement but clinicians currently do not have the tools to confidently adjust expectations. In this study, supervised machine learning was applied to multi-variate wearable sensor data from preoperative timed-up-and-go tests. Participants (n=82) were instrumented three months after surgery and patients showing relevant improvement were designated as ǣrespondersǥ while the remainder were labelled ǣmaintainersǥ. Support vector machine, naïve Bayes, and random forest binary classifiers were developed to distinguish patients using sensor-derived features. Accuracy, sensitivity, specificity, and area under the receiver-operator curve (AUC) were compared between models using ten-fold leave-subjects-out testing. A high performance using only sensor-derived functional metrics was obtained with a random forest model (accuracy = 0.76 ± 0.12, sensitivity = 0.85 ± 0.11, specificity = 0.60 ± 0.24, AUC = 0.82 ± 0.11) but a highly sensitive model was observed using a naïve Bayes model after including patient age, sex, and BMI into the feature set (accuracy = 0.73 ± 0.08, sensitivity = 0.98 ± 0.08, specificity = 0.32 ± 0.20, AUC = 0.80 ± 0.07). Including select patient-reported subjective measures increased the random forest performance slightly (accuracy = 0.81 ± 0.10, sensitivity = 0.93 ± 0.09, specificity = 0.61 ± 0.24, AUC = 0.86 ± 0.09). The current work has demonstrated that prediction models developed from preoperative sensor-derived functional metrics can reliably predict expected functional recovery following surgery and this can be used by clinicians to set realistic patient expectations.

Wearable sensors permit convenient human activity data collection in diverse environments and collected data can be used to evaluate functional impairment or analyze recovery following surgical interventions such as knee replacement. Automated activity classification can be used for adding context to unscripted sessions for comparing identical tasks across subjects. In this study, twenty participants were instrumented with wearable inertial sensors placed above and below both knees while performing activities of daily living. Collected multivariate time series data were encoded as colour images and three convolutional neural networks were developed to classify activities into eleven classes. Performance was evaluated using twenty iterations of a leave-one-subject-out scheme. A first-stage classification model was able to differentiate static vs. dynamic activities almost perfectly and a second-stage model was able to further classify specific static activities performed with 99% accuracy. A separate second-stage model was developed to classify dynamic activities with 91% accuracy. Cycling and ascending/descending stairs were the most commonly confused activities. The current work has demonstrated that both static and dynamic activities of daily living can be classified using only leg instrumentation which is beneficial for applications studying knee performance in varying environments.

Two novel nonlinear pose (i.e., attitude and position) filters developed directly on the Special Euclidean Group SE(3) able to guarantee prescribed characteristics of transient and steady-state performance are proposed. The position error and normalized Euclidean distance of attitude error are trapped to arbitrarily start within a given large set and converge systematically and asymptotically to the origin from almost any initial condition. The transient error is guaranteed not to exceed a prescribed value while the steady-state error is bounded by a predefined small value. The first pose filter operates based on a set of vectorial measurements coupled with a group of velocity vectors and requires preliminary pose reconstruction. The second filter, on the contrary, is able to perform its function using a set of vectorial measurements and a group of velocity vectors directly. Both proposed filters provide reasonable pose estimates with superior convergence properties while being able to use measurements obtained from low-cost inertial measurement, landmark measurement, and velocity measurement units. The simulation results demonstrate the effectiveness and robustness of the proposed filters considering large error in initialization and high level of uncertainties in velocity vectors as well as in the set of vector measurements.

This paper formulates the pose (attitude and position) estimation problem as nonlinear stochastic filter kinematics evolved directly on the Special Euclidean Group SE(3). This work proposes an alternate way of potential function selection and handles the problem as a stochastic filtering problem. The problem is mapped from SE(3) to vector form, using the Rodriguez vector and the position vector, and then followed by the definition of the pose problem in the sense of Stratonovich. The proposed filter guarantees that the errors present in position and Rodriguez vector estimates are semi-globally uniformly ultimately bounded (SGUUB) in mean square, and that they converge to small neighborhood of the origin in probability. Simulation results show the robustness and effectiveness of the proposed filter in presence of high levels of noise and bias associated with the velocity vector as well as body-frame measurements. Keywords: Pose estimator, pose observer, attitude estimate, estimator, observer, Nonlinear stochastic pose filter, stochastic differential equations, Brownian motion process, Ito, Stratonovich, Wong Zakai, special orthogonal group, homogeneous transformation matrix, complimentary filter, mapping, Parameterization, Representation, Robust, stability, uncertain, Gaussian, colored, white, noise, vectorial measurement, vector measurement, translational velocity, angular velocity, singular value decomposition, rotational matrix, identity, deterministic, comparison, inertial frame, rigid body, three dimensional, 3D, space, Lie group, projection, landmark, feature, gyroscope, micro electromechanical systems, Inertial measurement units, sensor, IMUs, Fixed, moving, orientation, Roll, Pitch, Yaw, SVD, UAVs, QUAV, unmanned, underwater vehicle, robot, robotic System, spacecraft, quadrotor, quadcopter, integral, comparative study, review, overview, autonomous, xyz, axis, SO(3), SE(3).