Together at Last: Perception and Learning for Collaborative Robots Operating in Human-Centric Environments

协作机器人或协作机器人是一类旨在与人类在共享空间中进行物理交互的机器。该定义涵盖了各种硬件系统和应用领域（从制造到自动驾驶汽车）。在接下来的五年里，这个新兴的细分市场预计将蓬勃发展，所有主要的机器人制造商都在这一领域推出了新产品。拟议研究计划的目标是大幅推进协作机器人技术的发展，使协作机器人比目前更有能力、更安全。这项研究将弥合差距，并有助于使协作机器人在实验室之外无处不在。该计划有四个关键研究主题：1）内省，传感器融合，传感器和执行器自校准，2）语义映射，3）基于信任的交互的意图识别，以及4）形式化测试和验证。这些主题通过高级感知处理的统一线索联系在一起，旨在为真实的世界创建灵活的协作机器人平台。四个主题（或项目领域）的研究活动将建立在机器人和机器学习的基础上。将开发一个层次感知的一般框架，从低级数据融合开始，并导致高级语义理解。来自新型传感模式的数据将被融合到一致的环境表示中;通过自我校准和内省，功能和安全的基线水平将始终保持不变。然后，多模态传感器数据将被处理以生成语义地图，其中将包括有关对象及其关系的信息（允许对交互进行推理）。这些地图将使用聚类和监督学习来制作。通过扩展语义感知，人类意图和活动识别将成为可能。意图识别和响应学习将显著提高安全性，确保每一方（人类和机器人）能够预测对方的行动。最后，所有功能将在广泛的测试活动中得到验证，包括使用先进的移动的机械手进行正式验证。拟议的研究计划有可能在广泛的行业产生变革性影响，从根本上改变人们与机器人的工作方式，进而利用机器人技术提高生产力和生活质量。研究成果将创造新的技术解决方案，这些解决方案将很容易转移到加拿大公司，为他们提供进入具有巨大增长潜力的新经济领域所需的工具。其中许多市场，包括服务机器人和先进制造业（例如，航空航天）将对加拿大在全球市场保持竞争优势至关重要。

Cobots, or collaborative robots, are a class of machines intended to physically interact with humans in a shared space. This definition encompasses a large variety of hardware systems and application domains (from manufacturing to autonomous vehicles). Over the next five years, this nascent market segment is expected to boom all major robotics manufacturers have established new products in this category. The goal of the proposed research program is to dramatically advance the state of the art in collaborative robotics, making cobots more capable and more safe than at present. This research will bridge the gap and help make cobots ubiquitous outside of laboratories. The program has four key research themes: 1) introspection, sensor fusion, and sensor and actuator self-calibration, 2) semantic mapping, 3) intention recognition for trust-based interaction, and 4) formal testing and validation. These themes are tied together by the unifying thread of advanced perceptual processing, with the aim of creating flexible cobot platforms for the real world.Research activities across the four themes (or project areas) will build upon results from robotics and machines learning. A general framework for hierarchical perception will be developed, beginning with low-level data fusion and leading to high-level semantic understanding. Data from novel sensing modalities will be fused into consistent environment representations; through self-calibration and introspection, a baseline level of functionality and safety will be maintained at all times. Multimodal sensor data will then be processed to generate semantic maps, which will include information about objects and their relationships (allowing for reasoning about interactions). The maps will be produced using clustering and supervised learning. By extending semantic perception, human intention and activity recognition will become possible. Intention recognition and response learning will significantly enhance safety, ensuring that each party (human and robot) is able to anticipate the actions of the other. Finally, all capabilities will be verified in an extensive test campaign that includes formal validation, using an advanced mobile manipulator.The proposed research program has the potential for transformative impact in a wide range of industries, fundamentally changing the way people work with robots and, in turn, leveraging robotics to enhance productivity and quality of life. Research outputs will create new technological solutions that will be readily transferable to Canadian companies, giving them the tools they need to enter new economic areas that have tremendous growth potential. Many of these markets, including service robotics and advanced manufacturing (for, e.g., aerospace) will be of critical importance to Canada in maintaining a competitive edge in the global marketplace.