教学课程资源库 更多>>

关键词： Physiology   Glucose   Neurons   Algorithms   Insulin   Knowledge discovery   Optimization   Parameter estimation   Biology   Computer science   Pharmaceutical sciences  

摘要： Type 1 diabetes is a chronic disease due to the patient’s pancreas producing no or little insulin. Unfortunately, there is no cure for type 1 diabetes. Controlling blood glucose in the euglycemic range is the main goal of developing an artificial pancreas for type 1 diabetes patients. The computer algorithm in the artificial pancreas can determine the time and amount of insulin delivery through the pump therapy. A computationally efficient and accurate model that can capture the physiological nonlinear dynamics is critical for developing an artificial pancreas. In this thesis, we find a few approaches to achieve the goal. Four data-driven models, including deep learning architecture and regression models, were employed for the blood glucose prediction using the real-world data obtained for 12 data contributors. A neural network model using an encoder-decoder architecture has the most stable performance and is able to recover missing dynamics in short time intervals. Regression models performed better at long-time prediction horizons (i.e., 60 minutes) and with lower computational costs. Next, we construct a complex physiological model to gain some interpretability for blood glucose dynamics. The physiological model contains six subsystems where the blood glucose transformation can be observable by solving the inverse problem and obtained the model parameters. Finally, we combine the data-driven model and physiological model for a hybrid model, which replaced one of the variables in the physiological model by reservoir computing, one of the regression models introduced in data-driven models. In this way, the hybrid model has the interpretability compared to data-driven models and is observed to improve the prediction accuracy compared to the physiological model.

关键词： Artificial intelligence   Energy   Computer science  

摘要： This thesis demonstrates the application of five machine learning regression models for predicting half-hour ahead solar irradiance and wind speed and compares the models’ performances. Historical meteorological data from 2016 to 2019 are collected from the National Renewable Energy Laboratory website for Pueblo, CO, USA. The dataset contains predictors such as temperature, pressure, humidity, wind speed, and solar irradiance for each half-hour of the day. The machine learning models include two models which support incremental learning. They are stochastic gradient descent and artificial neural network. The other three models are ridge regression, polynomial regression, and random forest regression. The significance of the predictors is calculated by using the point biserial correlation coefficient and the maximal information coefficient. The machine learning models are then trained with first 80% of the historical data with a range of hyperparameters. The performances of different hyperparameters are analyzed and the best ones are chosen to finally test the performances of the five machine learning models on last 20% of the data. R-squared, root mean square error, mean absolute error, and explained variance score are used as the performance metrics. Results show that all the models perform fairly close at predicting both at half-hour ahead solar irradiance and wind speed in terms of the performance metrics. However, the accuracy of wind speed prediction is higher than that of solar irradiance prediction. Furthermore, the computational times for the models are varied at a large scale, indicating that not all the models may be applicable in practical situations when making predictions every half-hour is required.

关键词： Computer science   Robotics   Artificial intelligence   Robot hands--Design and construction   Tactile sensors  

摘要： Robotic grasping and manipulation remains an active area of research despite significant progress over the past decades. Many existing solutions still struggle to robustly handle difficult situations that a robot might encounter even in non-contrived settings. For example, grasping systems struggle when the object is not centrally located in the robot's workspace. Also, grasping in dynamic environments presents a unique set of challenges. A stable and feasible grasp can become infeasible as the object moves; this problem becomes pronounced when there are obstacles in the scene. This research is inspired by the observation that object-manipulation tasks like grasping, pick-and-place or insertion require different forms of awareness. These include reachability awareness -- being aware of regions that can be reached without self-collision or collision with surrounding objects; tactile awareness-- ability to feel and grasp objects just tight enough to prevent slippage or crushing the objects; and 3D awareness -- ability to perceive size and depth in ways that makes object manipulation possible. Humans use these capabilities to achieve a high level of coordination needed for object manipulation. In this work, we develop techniques that equip robots with similar sensitivities towards realizing a reliable and capable home-assistant robot. In this thesis we demonstrate the importance of reasoning about the robot's workspace to enable grasping systems handle more difficult settings such as picking up moving objects while avoiding surrounding obstacles. Our method encodes the notion of reachability and uses it to generate not just stable grasps but ones that are also achievable by the robot. This reachability-aware formulation effectively expands the useable workspace of the robot enabling the robot to pick up objects from difficult-to-reach locations. While recent vision-based grasping systems work reliably well achieving pickup success rate higher than 90% in cluttered scen

关键词： Computer science   Artificial intelligence  

关键词： Computer science   Astronomy   Artificial intelligence   Aerospace engineering  

关键词： Artificial intelligence   Computer science  

摘要： Since more and more multi modality sensors are available to the consumers, one application of these sensors is to provide smart assistance in the kitchen environment for the better quality of lives. This thesis makes two contributions to provide the human-aware assistance in kitchen for average users: (i) Providing novel approaches for automatically detecting events of interest such as predicting food readiness and detecting liquid levels, (ii) Detecting user preferences such as predicting human grasp locations over the handle area of cups as the preliminary work, and providing multiple grasps on a single object. In this thesis, we are interested in contributing toward detecting the diversity of user preferences. We tackle that by first predicting multiple potential candidates for user preference on grasp on objects, to enable assistive robots that have knowledge on where people hold objects so as to deploy human-aware robot assistants in kitchens. Moreover, we propose two novel deep learning architectures for providing the better quality of multiple distinct grasps on objects in kitchens.

关键词： Unmanned aerial vehicles   Algorithms   Localization   Planning   Sensors   Aerospace engineering   Computer science   Engineering   Robotics   Transportation  

摘要： This dissertation develops a novel algorithm to perform autonomous surveys of unexplored 2D and 3D environments by extending the coverage path planning (CPP) literature to perform surveys using probabilistic coverage estimation. Robotic survey missions are designed to gather data from an environment and produce maps after the robot has been retrieved. The maps can take the form of 2D mosaics for planar terrain and 3D reconstructions for complex 3D environments. Survey missions of this form have applications in search-and-rescue, structural inspection, and scientific study. The goal of the survey mission is to achieve complete coverage, which is to have the robot observe every part of the environment before it is retrieved. Complete coverage becomes an issue when the robot does not have accurate localization, which can happen in environments that experience GPS-denial, e.g. underwater. In such environments, simultaneous localization and mapping (SLAM) is commonly used for localization. SLAM is an approach that corrects odometry with additional information in the form of terrain features. However, SLAM performance can degrade when there are insufficient features available, i.e. the terrain is feature-poor. There has been an evolution of algorithms that addresses the problem of surveying GPS-denied feature-poor terrain. State-of-the-art survey methods plan paths based on a coverage map, an estimate of which parts of the target area have been seen by the robot. This approach is called CPP. This thesis extends the CPP literature by developing an algorithm called SLAM directed by uncertainty-based node connections (SLAM-DUNC). For 2D surveys, SLAM-DUNC incorporates localization uncertainty into the coverage map to calculate aprobability of coverage and plans paths based on this quantity. This extension enables probabilistic coverage estimation with cameras. For 3D surveys, paths are planned using a similar quantity called the probability of occupancy. SLAM-DUNC was demons

关键词： Electrical engineering   Biomedical engineering   Computer science  

摘要： Diabetes is a major public health challenge affecting more than 451 million people. Physiological and experimental factors influence the accuracy of non-invasive glucose monitoring, and these need to be addressed before replacing the finger prick method with a non-invasive glucose measurement technique. Also, the suitable employment of machine learning techniques on experimental data can significantly improve the accuracy of glucose predictions. This work includes the design, development, testing and data analysis of an optical based sensor for glucose measurements. The feasibility of non-invasive measurement of glucose within aqueous solutions that assimilate the composition of human blood plasma is investigated. The laboratory testing of the sensor with controlled solutions helps to make valid conclusions about the performance of the sensor and the accuracy of the predictions. There are several goals associated with this study. The first goal is to use light sources with multiple wavelengths to enhance the sensitivity and selectivity of glucose detection in the aqueous solution. Multiple wavelength measurements have the potential to compensate for errors associated with inter- and intra-individual differences in blood and tissue components. In this study, the transmission measurements of the custom built optical sensor are examined using 18 different wavelengths between 410 and 940 nm. Results show a high correlation value (0.98) between glucose concentration and transmission intensity for 4 of the 18 wavelengths (485, 645 and 860 and 940 nm). The second goal of this study is to extract the intensity data using these 4 wavelengths and to collectively analyze the accuracy of glucose concentration predictions based on machine learning techniques. The intensity data measured using the four optimal wavelengths is first input into a support vector machine (SVM) classifier to discriminate hypoglycemic, normal and hyperglycemic ranges with 99% accuracy based on F1-scores

关键词： Islands   Aircraft   Embedded systems   Boolean   Open source software   Planning   Optimization   Robots   Algorithms   Pilots   Rain   Robotics   Aerospace engineering   Computer science   Engineering  

摘要： This thesis provides multi-robot path planning methods for conducting aerial surveys over large areas. In these surveys, Unpersoned Aerial Vehicles (UAVs) collect sets of images with their downward-facing cameras such that the entire area on the ground is covered. We present two novel methods for use in planning and deploying large-scale survey operations. The first method is an offline discrete path planning tool that finds a series of GPS waypoints for each UAV to visit such that the survey coverage requirements are met. The second is an online method that guarantees collision avoidance between UAVs and other objects in their environments despite noisy relative position uncertainty. Our path planning tool is designed to make the best use of limited flight time. Unlike current survey path planning solutions based on geometric patterns, our method, Path Optimization for Population Counting with Overhead Robotic Networks (POPCORN), solves a series of Boolean satisfiability instances of increasing complexity. Each instance yields a set of feasible paths at each iteration and recovers the set of shortest paths after sufficient time. By the addition of a divide-and-conquer scheme, Split And Link Tiles (SALT), these POPCORN instances can be computed in parallel and combined in a scalable manner to produce coverage paths over very large areas-of-interest. To demonstrate this algorithm's capabilities, we implemented our planning algorithm with a team of drones to conduct multiple photographic aerial wildlife surveys of the Cape Crozier Adélie penguin rookery on Ross Island, Antarctica, one of the largest penguin colonies in the world. The colony, which contains over 300,000 nesting pairs and spans over 2 square km, was surveyed in about 3 hours. In contrast, previous human-piloted single-drone surveys of the same colony required over 2 days to complete. We also have deployed our survey system in Mono Lake, CA to survey a California gull colony as well as a 2,000 acre ranch

关键词： Computer science   Artificial intelligence