Intelligent Coordination for Multi-Agent Systems

在许多现实世界的自治系统中，智能体团队必须在彼此之间分配任务，然后规划访问任务位置的轨迹。在拥挤和动态的环境中，代理可能必须避免接近人类和彼此，或者竞争资源，但为了一个共同的目标而合作。这种多智能体系统的应用包括自主飞机拖车、自动化仓库机器人、自动制导港口车辆、视频游戏中的游戏角色、巡逻机器人、搜救机器人、室内服务机器人、成群的无人地面和飞行器，以及其他多机器人系统。自主操作的协调是这些多智能体系统的基本构件，但随着智能体数量的增加和智能体的运行时间的增加，往往在计算上变得具有挑战性。因此，该研究计划旨在为做出高效、有效和健壮的决策建立算法基础，以协调这些多代理系统的自主操作。这项研究计划将研究不同的任务和运动协调问题，它们的复杂性，解决它们的算法，以及它们的应用。它还将展示我们的技术如何使用标准的机器人模拟器以及地面和空中机器人团队来帮助多代理系统的真实世界应用。具体地说，这项研究计划将涉及(1)发展理论洞察力，以表征计算复杂性并捕捉多智能体协调问题的新公式的组合结构，(2)通过利用这些问题的组合结构提供高效和有效的算法解决方案，以及(3)设计策略，即使在人类、系统动力学或不确定性存在的情况下，也可以将这些解决方案提炼成真实机器人的可执行计划。包括数百万辆自动驾驶汽车、自动无人机和服务机器人在内的智能代理团队很快将成为我们日常生活中不可或缺的一部分。我们相信，我们的技术将对协调和监管这些代理产生重大影响，并使它们变得更智能，以便更好地与人类合作和帮助人类。这项研究计划将在加拿大境内建立人工智能和机器人跨学科领域的专业知识。这也将有利于加拿大政府机构，这些机构旨在监管这些代理的地面和空中交通，并提高加拿大科技公司建立和使用智能代理团队的竞争力。算法技术将缩短这些智能体的运行时间，减少所需的智能体数量，同时保持系统的吞吐量不变，从而对污染、能源消耗、土地使用和拥堵产生积极的环境影响。

In many real-world autonomous systems, teams of agents have to assign tasks among themselves and then plan trajectories to visit the task locations. The agents may have to avoid proximity to humans and each other in congested and dynamic environments or compete for resources, yet working cooperatively toward a common objective. Applications of such multi-agent systems include autonomous aircraft towing vehicles, automated warehouse robots, automated-guided port vehicles, game characters in video games, patrolling robots, search-and-rescue robots, indoor service robots, swarms of unmanned ground and aerial vehicles, and other multi-robot systems. The coordination of autonomous operations is a fundamental building block for these multi-agent systems but often becomes computationally challenging as the number of agents and the operating time of agents increase. This research program thus aims to establish algorithmic foundations for making efficient, effective, and robust decisions to coordinate autonomous operations for these multi-agent systems. This research program will study different task and motion coordination problems, their complexities, algorithms for solving them, and their applications. It will also demonstrate how our techniques can benefit real-world applications of multi-agent systems using both standard robot simulators and teams of ground and aerial robots. Specifically, this research program will involve (1) developing theoretical insights that characterize the computational complexity and capture the combinatorial structure of novel formulations of multi-agent coordination problems, (2) providing efficient and effective algorithmic solutions to these problems by exploiting their combinatorial structure, and (3) designing strategies that refine these solutions into executable plans for real robots even in the presence of humans, system dynamics, or uncertainties. Teams of intelligent agents, including millions of self-driving cars, autonomous drones, and service robots, will soon become an intrinsic part of our daily life. We believe that our techniques will have a significant impact on coordinating and regulating these agents and making them smarter to better collaborate with and assist humans. This research program will build expertise in the interdisciplinary area of artificial intelligence and robotics within Canada. It will also benefit Canadian government agencies that aim to regulate ground and aerial traffic of these agents and enhance the competitiveness of Canadian technology companies that build and use teams of intelligent agents. The algorithmic techniques will shorten the operating time of these agents and reduce the number of agents needed, while keeping the same throughput of the system, thus resulting in a positive environmental impact on pollution, energy consumption, land usage, and congestion.