RI: Small: Collaborative Research: Learning to perform consistently in human/multi-robot teams

该项目侧重于在机器人可以探索不适合人类的地区的情况下，实际部署人类/多机器人团队。例如，一队“救援”机器人可以在倒塌的建筑物中搜寻受害者，并将他们的位置传递给外面的人类急救人员。在动态环境中管理一个人/多机器人团队是一个具有挑战性的问题。不仅游戏世界是多变的，团队成员也会发生变化，因为机器人迷路了，或者操作员需要休息——世界在变化，探索世界的团队也在变化。本研究的目标是为人类/多机器人团队制定策略，以学习一致和有效地执行。将追求三个主要目标：首先，通过一个实用的机构记忆框架来减轻团队组成的变化，该框架可以记住和使用过去的经验；第二，通过学习人类操作员的行为，为以后使用建模和记录专业知识；第三，通过提供一个平衡的社会选择机制，在团队成员之间有效地分配任务。这个项目的新方法适用于广泛的人类/多机器人和人类/多智能体团队，通过整合制度记忆，向人类队友学习，并解决不同观点之间的冲突。这些策略将使用一个人/多机器人试验台进行评估，该试验台由一名人类操作员和一组廉价、功能有限的异构机器人组成。虽然每个单独的机器人都有有限的机动性和传感能力，但团队成员共同构成了一个多功能的人/多机器人设施。该项目解决了鲁棒智能中的重要挑战，包括行为建模、从经验中学习、做出协调决策和不确定性下的推理。预期的结果包括人类/多机器人团队的策略，这些团队学习在各种条件下有效协作，并且可以在团队成员的运行时变化的情况下保持他们的表现，以及关于人如何与机器人团队互动的知识。更广泛的影响包括为远程合作者提供联网实验试验台的访问；出版针对研究生、本科生和高中生的多机器人团队的课程材料；让本科生参与研究活动；并与当地博物馆的现有联系人合作，向公众展示结果。

This project focuses on practical deployment of human/multi-robot teams in situations where robots can explore regions that are unsuitable for humans. For example, a team of "rescue" robots can sweep through a collapsed building searching for victims and transmit their positions to human first-responders outside. Managing a human/multi-robot team in a dynamic environment is a challenging problem. Not only is the world mutable, but also the team can experience altered membership because a robot gets lost or a human operator needs rest---the world is changing, and so is the team that is exploring that world.The goal of this research is to develop strategies for human/multi-robot teams to learn to perform consistently and effectively. Three primary aims will be pursued: first, to mitigate changes in team composition via a practical framework for institutional memory that remembers and uses past experiences; second, to model and record expertise for later use by learning behaviors performed by a human operator; and third, to distribute tasks among team members efficiently by providing a balanced mechanism for social choice. The novel approach of this project is applicable to a broad spectrum of human/multi-robot, and human/multi-agent teams, by integrating institutional memory, learning from human teammates, and resolving conflict among differing perspectives. The strategies will be evaluated using a human/multi-robot testbed comprised of one human operator plus a heterogeneous set of inexpensive, limited-function robots. Although each individual robot has restricted mobility and sensing capabilities, together the team members constitute a multi-function, human/multi-robot facility.This project addresses important challenges in robust intelligence, including behavior modeling, learning from experience, making coordinated decisions, and reasoning under uncertainty. Expected outcomes include strategies for human/multi-robot teams that learn to collaborate effectively under a variety of conditions and can maintain their performance despite run-time changes in team membership, as well as knowledge about how people interact with robot teams. Broader impacts include providing access to a networked experimental testbed for remote collaborators; publishing proven curricular materials on multi-robot teams addressed to graduate, undergraduate and high school students; involving undergraduates in research activities; and working with existing contacts at local museums to demonstrate results to the general public.