Trustable dexterous manipulation: morphologies and low-level control schemes for next-generation robot hand technologies

多指机械手和被机械手抓取和操纵的物体是机械手灵巧操作系统的组成部分。然后，灵巧操作问题可以定义为确定如何通过手指的协调运动改变对物体的抓取，以达到其位置和方向的期望变化的行为。在结构化环境中，也就是说，在其特征事先已知的世界中，解决灵巧操作问题可以简化为优化硬件和软件，以处理所寻求任务中存在的特定对象和约束。由于在这些情况下，所有可能的后果都被记录下来，设计优化通常得出的结论是，带有简单的两指颚或真空抓取器的多自由度机械臂足以定位和定向被操纵物体；从而避免了与实现机器人手灵巧性相关的困难。换句话说，根据所描述的分析，精细操作（指通过机器人手指、手和手腕等小机器人部件对物体的操作）被粗操作（指通过机器人手臂或其他类型的肢体等大机器人部件对物体的操作）所吸收。这当然是当前许多工业应用中观察到的典型情况，正如最近的协作工业机器人的设计所证明的那样。上述推理简单地解释了工业环境中缺乏使用多指灵巧机械手的原因，这是机器人操纵界最近讨论的一个主题，并清楚地揭示了为什么迫切需要这项技术和对灵巧操作问题的研究。这个问题的答案很简单：解决本世纪乃至未来一些最相关的社会、环境和经济挑战（例如，高效的医疗保健、应对人口老龄化、管理超大城市），需要机器人与人类合作，操纵为人类双手设计的物体。因此，考虑到这些设置固有的多样性和不确定性，机器人操作技术需要粗操作和精细操作的合作，而不是吸收。解决在非结构化环境中灵巧地操作对象的问题是必须的。然而，尽管机器人技术在过去的35-40年里取得了长足的进步，但在形状多样性和形状不确定性的情况下，用机器人手进行可靠的灵巧操作仍然是一个悬而未决的问题。本研究的目的是通过研究新的形态和低级控制方案来帮助解决这一问题，并塑造下一代机器人手技术，从而大大提高当前解决方案的灵巧操作能力。具体来说，本研究的重点是设计基于柔性和自适应机械部件的机器人手，该机械人手-物体系统产生重要的可预测行为，能够在开环中控制，即无需反馈控制，无需事先知道物体的特殊性，同时仍然对被操纵物体的大小或形状具有鲁棒性。这种新颖的方法被称为“可信赖的灵巧操作”，与传统的以手为中心的策略不同，它采用了一种整体的观点，既考虑了被操纵的身体，又不失一般性；它有可能重新定义当前灵巧机器人手设计的实践。该项目的成功将有利于研究机器人与人类在多个领域（包括农业、医疗保健、制造业和极端环境）在动态和不确定环境中合作的技术的研究人员和从业人员。

A multifingered robotic hand and an object that will be grasped and manipulated by the hand are the components of a dexterous manipulation robotic system. Then, the dexterous manipulation problem can be defined as the act of determining how to alter a grasp of an object through the coordinated motion of the fingers to reach a desired change in its position and orientation.In structured environments, that is, in worlds whose characteristics are well known in advance, solving the dexterous manipulation problem reduces to optimising hardware and software for dealing with the specific objects and constraints present in the sought task. Since in these cases all possible ramifications are documented, the design optimisation usually concludes that multi-degree-of-freedom robot arms with simple two-finger jaw or vacuum grippers are enough to position and orient the manipulated objects; thus avoiding the difficulties associated with implementing robot hand dexterity.In other words, as a result of the described analysis, fine manipulation, which refers to the manipulation of objects by small robot parts such as robotic fingers, hands, and wrists, is absorbed by gross manipulation, which refers to the manipulation of objects by large robot parts such as robotic arms or other types of limbs. This is certainly the typical situation observed in many of the current industrial applications, as demonstrated by the design of the most recent collaborative industrial robots.The above reasoning gives a simple explanation about the lack of use of multifingered dexterous robotic hands in industrial settings, an aspect that have been the subject of discussion in the robotic manipulation community recently, and clearly opens the question about why this technology and research on the dexterous manipulation problem is a pressing need.The answer of such a query is simple: the solution of some of the most relevant social, environmental, and economic challenges of this century, and beyond, (e.g., an efficient healthcare, coping with an ageing population, management of mega cities), requires robots that cooperate with humans to manipulate objects designed for human hands. Thus, given the diversity and uncertainty inherent of such settings, robot manipulation technologies require the cooperation, not the absorption, of gross manipulation and fine manipulation. Solving the problem of manipulating objects dexterously in unstructured environments is then a must.However, despite the substantial progress made in the last 35-40 years in robotics, performing reliable dexterous manipulation operations under both shape diversity and shape uncertainty with a robot hand is still an open question. The aim of this research is to help solving this problem and shaping the next generation of robot hand technologies by investigating novel morphologies and low-level control schemes that drastically enhance the dexterous manipulation capabilities of current solutions.Specifically, this research focuses on devising robot hands based on flexible and adaptive mechanical components that generate non-trivial predictable behaviours of the hand-object system that are able to be controlled in open loop, that is, without feedback control and without knowing the particularities of the object beforehand, while still being robust to the size or shape of the object being manipulated. This novel approach, called 'trustable dexterous manipulation', departs from traditional hand-centred strategies to embrace a holistic view that takes into account the manipulated bodies without losing generality; it has the potential to redefine the current practice in design of dexterous robot hands.The success of this project will benefit researchers and practitioners working on technologies that involve robots collaborating with humans in dynamic and uncertain settings across multiple domains, including agriculture, healthcare, manufacturing, and extreme environments.