NRI: FND: COLLAB: Coordinating Human-Robot Teams in Uncertain Environments

机器人硬件的成本下降和日益复杂，为机器人团队与熟练的人类组合部署创造了新的机会，以支持和加强劳动密集型和/或危险的体力劳动。机器人的愿景是将时间从熟练工人身上解放出来，这样他们就可以专注于自己擅长的任务(解决复杂问题、灵活操作、客户服务等)。机器人可以帮助处理工作中令人分心和令人沮丧的部分，比如运送材料或提取补给。这一愿景正在美国经济的许多部门和海外实现，如仓库管理、组装制造和灾难应对。然而，这一领域的进展正受到目前僵化和僵化的方法的阻碍，这些方法依赖于不切实际的人-机器人交互模型。这个项目试图通过提出团队机器人与团队人类协调完成复杂问题的新模型和方法来克服这些问题。特别是，该项目将创建和解决在不确定环境中协调人类和机器人团队的现实模型。PIS将探索这一研究领域的创新方法，并将做出以下贡献：1)实现对多人多机器人团队的变革性重新概念化，以准确地反映团队在时间动态、随机环境中的优势和局限性；2)开发考虑不确定性和部分可观测性的人-机器人团队的现实和通用模型；以及3)在这些模型中为规划和学习贡献创新和可扩展的技术。这项研究将建立在不确定和部分可观测的单机器人问题中取得成功的方法的基础上：部分可观测马尔可夫决策过程(POMDP)。POMDP模拟机器人和环境，但不是人类。然而，在几乎所有现实世界的应用中，明确地将人包括在这些模型中将是至关重要的。通过将POMDP扩展到与人类团队交互的多个机器人，可以表示混合人类和机器人团队的复杂和现实问题。在这个项目中开发的解决方法将允许机器人对领域和他们的人类队友的不确定性进行推理，同时优化他们的行为。这些方法广泛适用于人-机器人协作领域，但它们将在急诊室进行评估，这是一个具有大量不确定性的环境，在大容量时间内有许多交付和补给任务。一组机器人可以协助完成这些任务。实验将在模拟和加州大学圣地亚哥仿真培训中心进行，实验对象包括不同数量的人类和机器人。该项目的成果有可能改变人类与机器人进行协调的方式。

The decreasing cost and increasing sophistication of robot hardware is creating new opportunities for teams of robots to be deployed in combination with skilled humans to support and augment labor-intensive and/or dangerous manual work. The vision is for robots to free up time of skilled workers so they can focus on the tasks that they are skilled at (complex problem solving, dextrous manipulation, customer service, etc.) and robots can help with the distracting and frustrating parts of working, such as delivering materials or fetching supplies. This vision is being realized across many sectors of the US economy and abroad, such as in warehouse management, assembly manufacturing, and disaster response. However, progress in this area is being stymied by current methods that are rigid and inflexible, and rely on unrealistic models of human-robot interaction. This project seeks to overcome these problems by proposing new models and methods for teams robots to coordinate with teams humans to complete complex problems. In particular, this project will create and solve realistic models for coordinating teams of humans and robots in uncertain environments. The PIs will investigate innovative approaches to this research area, and will make the following contributions: 1) Enable a transformative re-conceptualization of multi-human multi-robot teamwork the accurately reflects the strengths and limitations of the team, as situated within a temporally dynamic, stochastic environment, 2) develop realistic and general models of human-robot teamwork that consider uncertainty and partial observability, and 3) Contribute innovative and scalable techniques for planning and learning in these models. This research will build off of methods that have been successful in single-robot problems under uncertainty and partially observability: partially observable Markov decision processes (POMDPs). POMDPs model robots and environments, but not humans. However, explicitly including people in these models will be critical in almost all real-world applications. By extending POMDPs to multiple robots interacting with teams of humans, complex and realistic problems with mixed human and robot teams can be represented. The solution methods developed in this project will allow the robots to reason about the uncertainty about the domain and their human teammates, while optimizing their behavior. The methods are broadly applicable to human-robot collaboration domains, but they will be evaluated in an emergency department, an environment with a large amount of uncertainty and many delivery and supply tasks during high-volume times. A team of robots can assist in these tasks. Experiments will take place in simulation and in the UC San Diego Simulation and Training Center with various numbers of humans and robots. The results of this project have the potential to transform the way human-robot coordination is performed.