Human-Robot Collaboration for Flexible Manufacturing

作为高复杂性、小批量产品的制造商，且构建标准差异很大，我们在制造过程中广泛采用机器人技术的障碍之一是所需的编程量 - 目前很难看到足够的投资回报。机器人能够理解与人类协作期间需要什么，并通过模仿学习自主生成解决方案，这将大大减少编程时间，从而实现敏捷、可重新配置的制造。当前关于人机交互以及人与协作机器人（协作机器人）之间协作的机器人和人工智能研究导致了意图识别的机器学习模型的设计，例如用于预测合作任务的目标（例如，Vinanzi 等人，2019）以及通过动觉模仿和语言指令进行动作学习（Zhong 等人，2019）。这些模型是为实验室环境中简化的联合任务场景而开发的。人工智能的最新进展（例如深度学习；Sunderhauf 等人，2018）提供了一个及时的机会来扩展有意阅读模型，以便在现实的行业环境场景中进行协作。这将导致系统了解其环境以及需要与之交互的组件，结合使用视觉和动作识别人工智能方法以及 CAD 模型/数字孪生的询问。博士研究任务：1。用于意图阅读和模仿学习的机器学习文献范围以及人机联合任务和人工智能方法的选择2.人机合作和学习任务的训练数据集生成3。对数据集的机器学习模拟以及人机交互实验的评估该项目与 EPSRC 在机器人和人工智能方面的优先领域直接相关。

As manufacturers of high complexity, low volume products with considerable variation in build standards, one obstacle to our adoption of widespread robotics in the manufacturing process is the amount of programming required - it is currently difficult to see sufficient return on investment. The ability for the robot to understand what is required during collaboration with people to produce a solution autonomously and via imitation learning, would drastically reduce programming times, resulting in agile, reconfigurable manufacturing. Current robotics and AI research on human-robot interaction and collaboration between people and cobots (collaborative robots) has led to the design of machine learning models of intention recognition, e.g. for the prediction of the goal of a cooperative task (e.g. Vinanzi et al. 2019) and for action learning via kinaesthetic imitation and linguistic instruction (Zhong et al. 2019). These models have been developed for simplified joint task scenarios, in laboratory settings. The latest advances in AI (e.g. deep learning; Sunderhauf et al. 2018) offer a timely opportunity to scale up intentional reading models for collaboration in realistic, industry setting scenarios. This will lead to the system understanding of its environment and the components with which it is required to interact, using a combination of visual and action recognition AI methods and interrogation of CAD models/digital twin.PhD research tasks:1. Scope of the literature on machine learning for intention reading and imitation learning and selection of human-robot joint task and of AI methods 2. Training dataset generation for the human-robot cooperation and learning task3. Machine learning simulations on dataset and a valuation in human-robot interaction experimentsThe project is directly aligned with EPSRC's priority areas in Robotics and AI.