Enhanced Robotic Gripper Optimisation: Simulation utilising Machine-Learning (ERGO:SuM)

机器人颗粒抓取器利用颗粒堵塞的过程来操纵各种各样的物体，代表了一种非常有前途的新兴技术，其潜在的适用性几乎涵盖了所有工业部门。虽然这些软机器人抓取器有可能对现代制造工艺产生变革性影响，但当前的系统仍然没有优化和不可靠，其微观动力学和结构几乎完全没有研究。该项目的目标是开发一种新的高度适应性的机器人抓取器系统，其功能可以快速自主优化，精确和安全地操纵各种物体。我们的工作将采取新的方法，将数值粒子模拟方法与人工进化算法相结合，通过虚拟原型开发新的设计，并校准和完善原型本身的操作，以最佳地处理显著不同的对象。此外，这些计算方法将用于驱动新的机器学习过程，使我们的系统能够“学习”如何最好地操纵新的“类”对象，从而在遇到属于熟悉类的先前未知对象时立即优化它们的行为。与涉及物理原型和“试错”实验和测试的传统技术相比，更容易且成本更低。我们独特的方法还使我们能够研究使用传统方法不可行的参数空间，并探索更冒险的，“高风险，高回报”的新设计和方法，而不必担心失败的代价。除了开发先进的技术用于颗粒抓具系统的设计、“训练”和优化之外，我们还将在模拟和物理两方面创建和测试新一代抓具，其强度和适应性通过使用能够产生局部颗粒流化和捕获的目标模式的内部压电换能器来增强。最后，我们还将进行第一次，使用最先进的三维成像技术，包括正电子发射粒子跟踪和X射线计算机断层扫描，对操作颗粒夹持器系统的内部动力学和结构进行深入的实验研究。所获得的详细信息将使我们能够将颗粒抓取器的宏观行为与其微观细节联系起来，从而使我们能够在基础层面上进一步理解这些重要系统。

Robotic granular grippers, which exploit the process of granular jamming to manipulate a vast range of diverse objects, represent a hugely promising nascent technology, with potential applicability spanning almost all industrial sectors. While these soft-robotic grippers carry the potential to prove transformative to the modern manufacturing process, current-generation systems remain unoptimised and unreliable, and their microscopic dynamics and structures almost entirely unresearched.The goal of this project is to develop a new breed of highly-adaptable robotic gripper systems whose function can be rapidly and autonomously optimised for the strong, precise and safe manipulation of a vast range of objects. Our work will take the novel approach of applying numerical particle simulation methods coupled with artificial evolution algorithms both to develop new designs via virtual prototyping, and to calibrate and refine the operation of the prototypes themselves in order to optimally handle significantly differing objects. Further, these computational methods will be used to drive novel machine-learning processes, enabling our systems to `learn' how best to manipulate new `classes' of objects and thus instantaneously optimise their behaviour when encountering a previously unknown object falling within a familiar class.The computational approach pioneered here will allow the aforementioned tasks to be performed more rapidly, more safely, more easily and at a reduced cost as compared to conventional techniques involving physical prototypes and `trial and error' experimentation and testing. Our unique methodology also enables us to investigate a parameter space that would be unfeasible using conventional methods, and to explore more venturesome, "high-risk, high reward" new designs and approaches without fearing the costs of failure. In addition to developing advanced techniques for the design, `training' and optimisation of granular gripper systems, we will also create and test - both in simulation and physically - a new generation of gripper whose strength and adaptability are enhanced via the use of internal piezoelectric transducers capable of producing targeted patterns of localised granular fluidisation and arrest.Finally, we will also conduct a first, in-depth experimental study concerning the interior dynamics and structure of operational granular gripper systems using state-of-the-art three-dimensional imaging techniques including Positron Emission Particle Tracking and X-ray computed tomography. The detailed information obtained will allow us to connect the macroscopic behaviours of granular grippers to their microscopic details, thus enabling us to further our understanding of these important systems on a fundamental level.