Search

Producing successful guide dogs for the visually impaired is a costly process and many dogs do not succeed in training. Identifying and screening successful dogs early is vital for the success of guide dog programs. Volunteer puppy raising is an essential part of training to expose candidate dogs to new stimuli and teach obedience skills away from the school. As there is little objective data about puppies' experiences with raisers, there is great potential in providing analytical insight during this important developmental period. We present a collar-based sensor system to remotely collect various sensor data to derive higher-level analytics describing behaviors and environments. We interviewed guide dog experts to refine and develop metrics derivable from sensor data. With the collar system, we collected training data and generated machine learning models to capture these metrics in order to remotely provide insight remotely at wide scale insight about puppies raised by volunteers.

As puppies progress through the training process to become guide dogs, they regularly undergo standard evaluations to assess their temperament. These evaluations consist of exposing the dogs to several stimuli while a highly-experienced observer scores them in several categories. In this paper, we present a neural network capable of generating the same scores as the human observer by using accelerometer and gyroscope data recorded during the evaluation. We define a baseline accuracy and compare the performance of the network, ultimately showing that it can achieve an accuracy score of approximately 93%.

This paper demonstrates the design and manufacturing of a smart and connected internet-of-things collar system for the collection of behavioral and environmental information from working canines. The environmental factors of ambient light, ambient temperature, ambient noise levels, barometric pressure and relative humidity are recorded by the smart collar system in addition to behavioral information about barking incidences and activity levels. The data are collected from the sensors and transmitted via Bluetooth to the handler’s smartphone where the custom app also acquires GPS positioning using the on-board smartphone sensors. The stored data on the smartphone are uploaded to the IBM Cloud once the user is connected to a WiFi network. The low power design of the smart collar system permits it to be used continuously for 27 hours with a 290 mAh lithium polymer battery. The cost of the system is low enough to let the handlers have multiple collars and exchange it if needed or recharge it overnight when not in use. This system is currently being scaled up to be tested on hundreds of canine puppies by a preeminent guide dog school in the US. As a result, the design emphasis here has been on the cost and power reduction, comfortable ergonomics, user friendliness, and robustness of data streaming. We expect the system to provide continuous quantitative data for improving guide dog training programs in addition to contributing the well-being of other working dogs in the future.

Evaluating potential guide dogs is crucial for guide dog schools as raising and training is an expensive process. During adolescence, volunteers raise dogs in training away from guide dog schools and expose them to a variety of stimuli and teach them obedience skills. However, no objective data exists about the dog's behavior and environment during this period, usually lasting several months to a year. We developed an Internet of Things sensor-equipped collar to quantify dogs' behaviors and environments during this stage. Raisers collect data from the collar using a smartphone app which in turn uploads data to a central processing pipeline. We present an overview of the system and an evaluation showing how we can learn meaningful information about a dog's environment and physical activities while away from the school for months on end, ideally to help predict which dogs will be successful in training.

In many of the applications involving working canines, such as search and rescue operations and agriculture, drones are increasing in ubiquity. There has been some recent focus on training dogs to follow or interact with drones in such applications. From the Animal Computer Interaction (ACI) perspective, drones have the potential to monitor the dog's welfare when handlers are not in close proximity in such harsh environments. Our on going work has focused on canine welfare through the use of on-body sensors for monitoring behavior, physiology, and the micro-environment dogs are in. We have also used these technologies to explore computer-assisted training of canines and more recently included drones to complement these to assess the macro-environment dogs are working in. This work discusses the challenges and opportunities we learned during our efforts to include drones for computer-assisted interactions with working dogs. We focus on the enabling technology and it's implications for ACI when dogs and drones work together.

Training a guide dog is a long and expensive process which involves experts with years of experience. At Guiding Eyes for the Blind, a large national guide dog school, a factor in the decision for whether a dog is suitable to continue training are numeric scores based on a subjective judgement during observation of the dog as it undergoes formal evaluations. As a step towards a more objective system, we outfitted dogs undergoing these evaluations with a data collection system capable of collecting electrocardiography and other data. Using both a prototype network and an optimized network, we show that electrocardiography data can be used to predict 29 behavioral scores with approximately 92% accuracy over 11 distinct tasks during the evaluation. Additionally, we show that each of the 11 tasks can predict any of the scores, indicating that the most predictive features in the data may be task agnostic.

This paper reports the development of a wearable wireless canine harness for tracking heart rate, heart rate variability, and activity level in working dogs. This new design is based on 3D-printed conductive electrodes embedded into the structure of a chest strap instrumented with a wireless data acquisition system. We tested the functionality of this system on adult detection dogs for their wellness assessment. The preliminary in vivo results demonstrate the feasibility of this system to track performance and wellness of working dogs up to walking speeds of 5-10 km/h.

Multi-axis mechanical stress sensors have become an important tool in numerous fields. To characterize the quality and resolution of a new stress sensor, researchers use test platforms to apply controlled amounts of shear and normal stresses. Since commercially available multi-axis test platforms are expensive, in many cases researchers end up needing to build their own sensor characterization platforms. In this study, we describe a cost-effective multi-axis stress sensor characterization platform to reduce development time, enable accurate benchmarking and provide an open source standard for stress sensor characterization for other researchers. In our design, we combined a 3-axis translational stage and a 3-axis force and torque (F/T) sensor through custom hardware and software. Our platform is cost effective and can reach the limits of the F/T sensor (Fxy = ±65 N and Fz = ±200 N) without losing any accuracy to structural bending. We provide detailed construction and operation instructions as well as results of extensive static and cyclic load experiments. To further validate our system, we show characterization results from a custom stress sensor as a case study. The modular nature of our platform also enables other researchers to customize this characterization platform for their unique experimental requirements. Keywords: Mechanical characterization, Stress sensor design, Testing stage, Stress sensor calibration

Animal-assisted therapies (AAT) are becoming increasingly common to help hospitalized patients, especially in oncology units. There is a critical need for methods and technologies that can enable a quantifiable understanding of AAT to objectively demonstrate its efficacy and improve its efficiency. In this paper, we present our preliminary efforts towards the development of wireless sensor systems to simultaneously detect the related behavioral (activity level, movement, stroking) and physiological signals (heart rate/variability) of humans and animals during their interaction. To detect heart rate, we tested two different techniques based on wearable or contactless electrocardiography. In this preliminary evaluation, we were able to assess these parameters successfully and identify the design challenges towards deployment of these systems in larger clinical studies.

This paper describes the design and fabrication of 3D-printed conductive electrodes for measuring heart rate and heart rate variability in animals. The customizable electrodes have a three-legged stool structure with rounded edges in order to provide optimized, balanced, and comfortable skin contact through the fur of the animal without needing to shave it. We explored two alternative designs for manufacturing: a flexible, insulated base structure coated with graphene for conductivity and a rigid, all-conductive, graphene-infused PLA base structure. To enable connection to standard female electrocardiogram snap connectors, we epoxied the former electrode with a metal male snap connector and the latter electrode had the standard-sized male snap connector 3D printed in one assembly. We characterized and compared the performance of these electrodes through electrochemical impedance spectroscopy and benchmarked with commercial electrodes traditionally used in veterinary clinics. Preliminary in vivo results demonstrate the feasibility of these electrodes to measure heart rate and heart rate variability in canine puppies.

Training successful guide dogs is time and resource intensive, requiring copious professional and volunteer labour. Even among the best programs, dogs are released with attrition rates commonly at ...

It is incredibly expensive to breed, train, and place a working guide dog; costs are tens of thousands of dollars for many organizations. During this training process, approximately 40% of canines are dropped due to medical, behavioral, or other reasons. Since guide dog schools do not have unlimited resources, they must focus on the most promising candidates so they can maximize resources to meet the growing need for guide dogs. Staff at some schools use evaluations to predict how successful a dog will be in the program. While current processes have achieved some positive results, they are highly-subjective, and rely on people with years of experience and knowledge. As a major first step towards an objective evaluation, we developed a system for a human operator to annotate canine behavior while simultaneously collecting physiological and accelerometer data. In collaboration with Guiding Eyes for the Blind, we designed the system from the ground up to meet the needs of both canine and human users.

Flow splitting to a mass spectrometer is a common way of coupling a highly specific detector to preparative or semi-preparative high-performance liquid chromatography (HPLC) purification of combinatorial libraries, drug metabolites, and characterizable impurities. The sensitive mass spectrometer consumes only a small fraction of the analyte while providing online structure-specific detection, and its output can thus be used to trigger collection of the desired fraction. Coupling mass spectrometry to preparative HPLC is difficult due to the susceptibility of the detector to fouling under conditions of high analyte concentration or solute amount, or to changes in solvent composition. We report here on a device, the mass rate attenuator (MRA), which automatically produces split ratios over a range of 100:1 to 100 000:1 under programmable user control. The MRA is a flow-control device that periodically gates a small aliquot from one liquid stream into another. The design allows the user to set the frequency of the gating without interruption of the HPLC flow stream. The MRA also allows control of the volume of the aliquot that is transferred between the flow streams. This additional control, compared to passive splitting devices, facilitates optimization of the tubing connecting the separation, detection and collection events. We demonstrate that such optimization can reduce the volume of the collected fraction without compromising recovery, thus reducing the time spent in evaporating solvents to reclaim purified products.

Patient and physician awareness and acceptance of trials and patient ineligibility are major cancer clinical trial accrual barriers. Yet, trials are typically conceived and designed by small teams of researchers with limited patient input. We hypothesized that through crowdsourcing, the intellectual and creative capacity of a large number of researchers, clinicians, and patients could be harnessed to improve the clinical trial design process. In this study, we evaluated the feasibility and utility of using an internet-based crowdsourcing platform to inform the design of a clinical trial exploring an antidiabetic drug, metformin, in prostate cancer. Over a six-week period, crowd-sourced input was collected from 60 physicians/researchers and 42 patients/advocates leading to several major (eg, eligibility) and minor modifications to the clinical trial protocol as originally designed. Crowdsourcing clinical trial design is feasible, adds value to the protocol development process, and may ultimately improve the efficiency of trial conduct.