RI: Small: To Ask or Not to Ask - A Foundation for the Optimization of Human-Robot Networks

无人驾驶车队在灾难救援、应急响应、国家安全等领域可以产生巨大的影响。人类操作员是这些任务中不可或缺的一部分。考虑到任务的复杂性、操作环境的严酷性和资源限制，对人类和机器人如何最好地交互的基本理解对于成功的操作至关重要。本文提出了一种新的多学科方法，用于在存在不确定性的情况下需要视觉搜索的任务上进行高效的人机协作。所提出的描述人类视觉表现的方法也有助于人类感知和认知心理学领域。该提案还针对加州中部地区和加州大学圣巴巴拉分校的代表性不足的学生提供了重要的教育内容。提出了一种新的方法来预测人们是否能对给定的感官输入做出正确的视觉决策。该方法基于训练卷积神经网络，使用来自亚马逊机械土耳其工人的大量输入。这种对人类视觉表现的基本理解将对机器人的野外决策产生重大影响。一套新的数学工具将开发用于机器人现场操作，给予人类视觉能力的正确理解。这使机器人能够正确地决定何时寻求人类帮助，何时依靠自己，以及何时更多地感知。最后，由于链路质量的空间变化性质，通信成本具有特征，这可以显著影响寻求人工帮助。提出的理论和设计范例将使用机器人试验台进行验证。

A team of unmanned vehicles can have tremendous impact in areas such as disaster relief, emergency response, and national security. Human operators are an integral part of these missions. Given the complexity of the tasks, harshness of the operation environments, and resource limitations, a fundamental understanding of how humans and robots can best interact is crucial to a successful operation. This proposal introduces a new multidisciplinary approach for efficient human-robot collaboration on tasks that require a visual search in the presence of uncertainty. The proposed methodology for characterizing human visual performance also contributes to the field of human perception and cognitive psychology. This proposal also has a significant educational component targeting under-represented students of the central California area and UC Santa Barbara.A new methodology is proposed for predicting whether people can make a correct visual decision for a given sensory input. The approach is based on training convolutional neural networks using extensive input from Amazon Mechanical Turk workers. This foundational understanding of human visual performance will have significant implication for robotic field decision making. A new set of mathematical tools will be developed for robotic field operation, given a proper understanding of human visual capabilities. This enables a robot to properly decide when to seek human help, when to rely on itself, and when to sense more. Finally, the cost of communication is characterized, due to the space-varying nature of link quality, which can significantly affect seeking human help. The proposed theories and design paradigms will be validated using a robotic testbed.