Dynamic Heterogeneous Multi-Robot Team Management in Dangerous Domains

在危险环境中工作，例如城市搜索和救援（USAR，灾后探索和结构测绘、寻找受害者和协助灾难响应）对智能机器人团队来说是一项极大的挑战。 世界是复杂且不可预测的，机器人可能会被损坏或摧毁，通信也会因基础设施受损而受到干扰。 需要由各种类型的机器人组成的团队（异质团队），因为通过更多的可消耗个体可以更好地分摊损失单位所涉及的风险。在这种情况下，团队必须围绕随时可用的人员组合动态地组建和重组，利用可用的技能做有用的工作，积极寻找新的人员来补充技能，并在可行的程度上利用受损的人员（包括在可能的情况下协助这些人员）。 个人必须同样适应伤害（例如，如果不再移动，则使用抓握肢体）并尽可能与团队合作。 贯穿所有这一切，风险必须与团队执行任何任务的价值积极平衡，并且这些任务都必须在快速变化和有限的沟通下完成。 这里提出的研究计划将推进危险环境中动态异构机器人团队合作的最先进技术，合并和扩展所有上述主题。 这些主题的工作已经单独完成（例如，更换团队的任务分配、招募代理），但支持像 USAR 这样危险的领域的广泛机器人集合的可部署去中心化解决方案仍然难以实现。 在最近的工作中，我和我的学生为此开发了一个框架，包括在改变团队成员的情况下完成任务、适应机器人的损失以及招募新人。 我们还一直致力于改进机器人控制和复杂领域的规划。 我将利用这一经验更广泛地推进动态异构团队合作，涵盖主要问题，例如尽可能管理受损机器人、协助受损机器人、平衡任务风险与价值以及制定管理风险的团队策略。 其结果将是一个软件系统，将在现实世界中使用机器人团队进行物理演示。 我们将探索人形机器人的使用，特别是在这些条件下的团队中，因为这种形式的性质尽管受到损害，但仍支持广泛的活动（例如，如果无法行走，人形机器人可以爬行或拖动自己）。 这项工作适用于美国陆军以外的广泛危险领域，例如采矿、太空探索和国防。 拟议的研究将为四名博士生和九名硕士生提供人工智能、复杂软件开发和机器人技术方面的宝贵培训：行业急需的技能。

Work in dangerous environments such as Urban Search and Rescue (USAR, the post-disaster exploration and mapping of structures, searching for victims and aiding disaster response) represents an extreme challenge to teams of intelligent robots. The world is complex and unpredictable, robots can be damaged or destroyed, and communication is interfered with due to damaged infrastructure. Teams consisting of various types of robots (heterogeneous teams) are required, because the risk involved in losing units can be better amortized with larger numbers of more expendable individuals. Under such conditions, teams must form and reorganize dynamically around the mix of individuals available at any time, doing useful work with the skills available, actively looking for new individuals to supplement skills, and making use of damaged individuals to the degree practicable (including assisting these if possible). Individuals must similarly adapt to damage (e.g. make use of grasping limbs if no longer mobile) and work with a team as well as possible. Permeating all of this, risk must be actively balanced with the value of carrying out any task for the team, and these tasks must all be accomplished under rapid change and limited communication. The program of research proposed here will advance the state of the art in dynamic heterogeneous robotic teamwork in dangerous environments, incorporating and extending all of the above themes. Work has been done in these themes individually (e.g. task allocation on changing teams, recruiting agents) but a deployable decentralized solution supporting a broad collection of robots for a domain as dangerous as USAR is still elusive. In recent work my students and I have developed a framework for this purpose that includes accomplishing tasks under changing team membership, adapting to robot loss and recruiting new individuals. We have also been working on improved robot control and planning for complex domains. I will be leveraging this experience to more broadly advance dynamic heterogeneous teamwork, encompassing major issues such as managing damaged robots to the best extent possible, assisting damaged robots, balancing task risk vs. value, and developing team strategies to manage risk. The outcome of this will be a software system that will be physically demonstrated using teams of robots in the real world. We will be exploring the use of humanoid robots in particular on teams under these conditions, since the nature of this form supports a broad range of activities despite damage (e.g. humanoids can crawl or drag themselves if unable to walk). This work is applicable in a broad range of dangerous domains beyond USAR, such as mining, space exploration, and defense. The proposed research will provide valuable training for four PhD students and nine MSc students in artificial intelligence, complex software development, and robotics: skills that are vitally needed by industry.