Learning-Aided User Intent Estimation for Smart Rollators.

With the aging population and rising rates of mobility disability, the demand for advanced smart rollators is increasing. To design control systems which improve safety and reliability, accurate prediction of human intent is required. In this paper, we present a classification method to predict intent of the rollator user using indirect inputs. The proposed classification algorithm uses data collected from an inertial measurement unit and an encoder implemented into a rollator. The developed intent estimation method is experimentally verified on our modified robotic platform. For our experiment with 7 healthy young adults, KNN classification algorithm was able to predict 3 intents (turn left, turn right and walk straight) with 92.9 % accuracy.

µPump: An open-source pressure pump for precision fluid handling in microfluidics.

An open-source precision pressure pump system and control software is presented, primarily designed for the experimental microfluidics community, although others may find additional uses for this precision pressure source. This mechatronic system is coined 'µPump,' and its performance rivals that of commercially available systems, at a fraction of the cost. The pressure accuracy, stability, and resolution are 0.09%, 0.02%, and 0.02% of the full span, respectively. The settling time to reach 2 bar from zero and stabilize is less than 2 s. Material for building a four-channel µPump (approx. $3000 USD) or an eight-channel µPump (approx. $5000 USD) is approximately a quarter, or a third of the cost of buying a high-end commercial system, respectively. The design rationale is presented, together with documented design details and software, so that the system may be replicated or customized to particular applications. µPump can be used for two-phase droplet microfluidics, single-phase microfluidics, gaseous flow microfluidics and any other applications requiring precise fluid handling. µPump provides researchers, students, and startups with a cost-effective solution for precise fluid control.

An integrated microfluidic device for the sorting of yeast cells using image processing.

The process of detection and separation of yeast cells based on their morphological characteristics is critical to the understanding of cell division cycles, which is of vital importance to the understanding of some diseases such as cancer. The traditional process of manual detection is usually tedious and inconsistent. This paper presents a microfluidic device integrated with microvalves for fluid control for the sorting of yeast cells using image processing algorithms and confirmation based on their fluorescent tag. The proposed device is completely automated, low cost and easy to implement in an academic research setting. Design details of the integrated microfluidic system are highlighted in this paper, along with experimental validation. Real time cell sorting was demonstrated with a cell detection rate of 12 cells per minute.

Development of a patient-specific anatomical foot model from structured light scan data.

The use of anatomically accurate finite element (FE) models of the human foot in research studies has increased rapidly in recent years. Uses for FE foot models include advancing knowledge of orthotic design, shoe design, ankle-foot orthoses, pathomechanics, locomotion, plantar pressure, tissue mechanics, plantar fasciitis, joint stress and surgical interventions. Similar applications but for clinical use on a per-patient basis would also be on the rise if it were not for the high costs associated with developing patient-specific anatomical foot models. High costs arise primarily from the expense and challenges of acquiring anatomical data via magnetic resonance imaging (MRI) or computed tomography (CT) and reconstructing the three-dimensional models. The proposed solution morphs detailed anatomy from skin surface geometry and anatomical landmarks of a generic foot model (developed from CT or MRI) to surface geometry and anatomical landmarks acquired from an inexpensive structured light scan of a foot. The method yields a patient-specific anatomical foot model at a fraction of the cost of standard methods. Average error for bone surfaces was 2.53 mm for the six experiments completed. Highest accuracy occurred in the mid-foot and lowest in the forefoot due to the small, irregular bones of the toes. The method must be validated in the intended application to determine if the resulting errors are acceptable.

Image processing and classification algorithm for yeast cell morphology in a microfluidic chip.

The study of yeast cell morphology requires consistent identification of cell cycle phases based on cell bud size. A computer-based image processing algorithm is designed to automatically classify microscopic images of yeast cells in a microfluidic channel environment. The images were enhanced to reduce background noise, and a robust segmentation algorithm is developed to extract geometrical features including compactness, axis ratio, and bud size. The features are then used for classification, and the accuracy of various machine-learning classifiers is compared. The linear support vector machine, distance-based classification, and k-nearest-neighbor algorithm were the classifiers used in this experiment. The performance of the system under various illumination and focusing conditions were also tested. The results suggest it is possible to automatically classify yeast cells based on their morphological characteristics with noisy and low-contrast images.

Experimental/analytical analysis of human locomotion using bondgraphs.

A new vectorial bondgraph approach for modeling and simulation of human locomotion is introduced. The vectorial bondgraph is applied to an eight-segment gait model to derive the equations of motion for studying ground reaction forces (GRFs) and centers of pressure (COPs) in single and double support phases of ground and treadmill walking. A phase detection technique and accompanying transition equation is proposed with which the GRFs and COPs may be calculated for the transitions from double-to-single and single-to-double support phases. Good agreement is found between model predictions and experimental data obtained from force plate measurements. The bondgraph modeling approach is shown to be both informative and adaptable, in the sense that the model resembles the human body structure, and that modeled body segments can be easily added or removed.