Real-Time Motion Planning for Complex Robotic Tasks

机器人正在走出工厂车间，进入仓库、私人住宅、商业工作场所、医院和户外。这种转变是由机器人技术的快速进步推动的。我们正在达到一个可以低成本购买地面车辆和空中无人机的地步。此外，当预先知道机器人环境时，就存在有能力的算法来规划执行复杂任务的运动。然而，现实世界的环境以不可预测的方式发展。人类可能在环境中移动，障碍物的位置可能未知，机器人甚至可能不确定它们的确切位置。这样的环境构成了巨大的挑战。机器人不能再完全提前规划它们的运动，它们必须安全和实时地对环境变化做出反应。 拟议研究的长期目标是开发灵活的机器人运动规划器，使机器人能够在面临不确定性的情况下高效地完成复杂的运动任务。这项研究计划背后的驱动思想是利用广泛的离线计算来确定运动策略和意外情况，为机器人的在线和实时执行做好准备。在网上，我们可以利用增量式运动规划器来高效地适应机器人的运动。 这项研究计划将解决与不确定性下的规划相关的几个关键研究挑战。这将包括在线修改运动计划的方法，自动确定机器人任务的哪些方面可以独立规划的方法，以及当机器人甚至不确定自己在工作空间中的位置时的规划技术。我们还将解决为具有复杂动力学的机器人寻找有效的运动计划的问题。 采用机器人技术的一个主要障碍是开发高效和安全的机器人运动控制器所需的初始投资。这项研究将在机器人运动规划方面取得根本性进展，帮助降低这些成本。在全球范围内，到2025年，机器人产业预计将增长到每年2000亿美元。加拿大的机器人产业正在迅速发展，随着这些公司的发展，他们正在将重点从制造机器人硬件过渡到提供完整的机器人解决方案。拟议的研究将为建造通用机器人以及提供完整的机器人解决方案奠定基础。该计划将在快速增长的机器人领域培养出高素质的人才。这些学生将有机会接触滑铁卢最先进的机器人设施，并将成为开发机器人自主性和使用机器人硬件方面的专家。这将使HQP处于一个不断增长和具有变革性的领域的前沿。

Robots are moving beyond the factory floor and into warehouses, private homes, commercial workplaces, hospitals, and the outdoors. This transition is being driven by rapid advances in robot technology. We are reaching a point where ground vehicles and aerial drones can be purchased at low cost. In addition, when the robot environment is known beforehand, capable algorithms exist for planning motion to perform complex tasks. However, real-world environments evolve in unpredictable ways. Humans may be moving in the environment, locations of obstacles may be unknown, and robots may not even be certain of their exact location. Such environments pose significant challenges. Robots can no longer fully plan their motion in advance, and they must react to environmental changes safely and in real-time. The long-term objective of the proposed research is to develop flexible robot motion planners that enable robots to efficiently complete complex motion tasks in the face of uncertainty. The driving idea behind this research program is to utilize extensive offline computation to determine motion strategies and contingencies, "preparing" the robot for the online and real-time execution. Online, we can leverage incremental motion planners to efficiently adapt robot motion. This research program will address several key research challenges related to planning under uncertainty. This will encompass methods for revising motion plans online, approaches for automatically determining what aspects of a robot mission can be planned independently, and techniques for planning when a robot is not even certain of its own location in its workspace. We will also tackle problems in finding efficient motion plans for robots with complex dynamics. A major barrier to the adoption of robotics is the initial investment required to develop efficient and safe robot motion controllers. This research will provide fundamental advances in robot motion planning that help reduce these costs. Worldwide, the robotics industry is expected to grow to $200 billion dollars annually by 2025. Canada's robotics industry is growing rapidly, and as these companies grow, they are transitioning their focus from building robot hardware to providing complete robotic solutions. The proposed research will lay the foundation on which general-purpose robots can be built, and from which complete robotic solutions can be offered. The program will produce highly qualified personnel in the rapidly-growing field of robotics. These students will have access to state-of-the-art robotics facilities at Waterloo, and will become experts both in developing robot autonomy and in working with robot hardware. This will position HQP at the forefront of a growing and transformative field.