NRI: Small: A Co-Robotic Navigation Aid for the Visually Impaired

描述（由申请人提供）：该项目的目标是为视障人士开发一种协同机器人导航辅助设备，称为协同机器人手杖（CRC）。CRC能够通过直观的人机交互（HDl）机制与用户协作，在3D环境中进行有效导航。CRCs导航功能包括设备位置估计、寻路、障碍物检测/避免和目标识别。使用儿童权利中心可改善视障人士的独立行动能力，从而改善他们的生活质素。该提案的教育计划是让研究生、本科生和高中生参与项目，并利用项目的活动来招收和留住工程专业的学生。该提案的智力优点是开发了新的计算机视觉方法，支持在3D环境中精确的盲导航和直观的HDl接口，以有效地使用设备。这些方法包括：(1)一种新的机器人姿态估计方法，该方法将自我运动估计和视觉特征跟踪相结合，提供准确的设备姿态；(2)基于高斯混合模型的模式识别方法，该方法可以识别室内结构和物体，用于寻路和避障操作；(3)一种创新的机制，可以直观地传递所需的行进方向；(4)用于设备模式自动切换的人的意图检测接口。计算机视觉方法的GPU实现将使实时执行成为可能。提出的盲航解决方案将赋予CRC先进的导航功能，而这些功能目前在现有的盲航辅助设备中是不具备的。PI的团队已经对计算机视觉方法进行了概念验证研究，结果很有希望。更广泛的影响包括：(1)这些方法的短期应用将影响广大视障人士；(2)这些方法将提高小型机器人和便携式机器人设备的自主性，这些设备在军事监视、执法和搜索和救援中有着无数的应用；(3)该项目将改善Pi大学的研究基础设施，并为未来的科学和工程职业培养研究生和本科生。

DESCRIPTION (provided by applicant): The project's objective is to develop enabling technology for a co-robotic navigation aid, called a Co-Robotic Cane (CRC), for the visually impaired. The CRC is able to collaborate with its user via intuitive Human-Device Interaction (HDl) mechanisms for effective navigation in 3D environments. The CRCs navigational functions include device position estimation, wayfinding, obstacle detection/avoidance, and object recognition. The use of the CRC will improve the visually impaired's independent mobility and thus their quality of life. The proposal's educational plan is to involve graduate, undergraduate and high school students in the project, and use the project's activities to recruit and retain engineering students. The proposal's Intellectual Merit is the development of new computer vision methods that support accurate blind navigation in 3D environments and intuitive HDl interfaces for effective use of device. These methods include: (1) a new robotic pose estimation method that provides accurate device pose by integrating egomotion estimation and visual feature tracking; (2) a pattern recognition method based on the Gaussian Mixture Model that may recognize indoor structures and objects for wayfinding and obstacle manipulation/ avoidance; (3) an innovative mechanism for intuitive conveying of the desired travel direction; and (4) a human intent detection interface for automatic device mode switching. The GPU implementation of the computer vision methods will make real-time execution possible. The proposed blind navigation solution will endow the CRC with advanced navigational functions that are currently unavailable in the existing blind navigation aids. The PI's team has performed proof of concept studies for the computer vision methods and the results are promising. The broader impacts include: (1) the methods' near term applications will impact the large visually impaired community; (2) the methods will improve the autonomy of small robots and portable robotic devices that have a myriad of applications in military surveillance, law enforcement, and search and rescue; and (3) the project will improve the research infrastructure of the Pi's university and train graduate and undergraduate students for their future careers in science and engineering.