Improved robotic locomotion performance through morphological computation and active control

多年来，具有腿部运动的机器人表现出高度的灵活性、良好的动态稳定性和对不同地形和障碍物的适应性[1]。他们在不同的环境中展示了这些特征，特别是在救援、侦察、保健、安全和海洋环境等领域[2]。机器人已被证明在涉及阴暗、肮脏和危险环境的应用中具有优势，而核退役等关键任务仍然是国家的关键优先事项。人们普遍认为，超过50%的地球表面是车轮或轨道无法到达的[2]，从最近的商业成功(如ANYbotics[3])以及成本效益高的开源系统(如Stanford Doggo[4]和ODRI[5])来看，腿部机器人发挥着越来越大的作用。尽管腿部机器人的发展展示了安全性、健壮性和高性能，但很明显，商业上可用的机器人是通过集中控制设计的，每个关节都是驱动的。这里，有必要介绍形态计算(MC)，在具体化(人工)智能的背景下，它指的是由身体(和环境)进行的过程，否则必须由大脑来执行[6]。MC在生物和机器人系统的研究中是相关的，如被动动态步行器[7]所说明的，这是一个纯粹的机械系统。被动动态步行器表明，步行可以是系统的物理特性和环境相互作用的结果，而不需要驱动。在机器人学的背景下，这意味着具有高度形态计算的系统只需要在需要时生成电机命令。这样的控制方案不仅增加了系统的耐用性(因为执行器的磨损减少了)，还意味着具有高MC的机器人对其驱动的能量需求将会减少[8]。这对有腿的机器人很有用，因为它们需要解开绳索才能自主运作，并在具有挑战性的地形上运输有效载荷。虽然MC在生物系统中的作用得到了很好的接受，但Ghazi-Zahedial[8]在他们对最近趋势的探索中表示，它在机器人学中的应用仍未得到充分探索。作者强调的主要原因之一是，传统的控制模式将身体视为需要被控制的东西，而不是被用作计算资源。有一种趋势是通过使用控制系统和伺服电机来抑制任何不受欢迎的形态行为，如非线性、欠驱动或噪声。值得注意的是，这些相同的复杂的形态特征在自然系统的行为中扮演着关键角色，正如阿巴德等人在MC中所描述的那样[9]和小狗[7]，一个欠驱动的机器人。同时，Deimal等人[10]提出的挑战之一是确保系统利用形态(“良好MC”)，同时避免与所需功能相关的有害的身体-环境相互作用(“不良MC”)。为了实现MC和主动控制之间的平衡，需要一种形式化的方法来测量系统中的MC。在详细介绍算法的论文中，Ghazi-Zahedi等人[6]证明了两种度量MC的方法，其中一种是比较行为复杂性和控制器复杂性，前者通过世界状态的信息来比较，后者通过传感器状态的信息来比较。回顾学术界和工业界的现状，得出的结论是，利用计算能力、集中控制和形态特征的相互作用的机器人系统具有巨大的潜力，可以在不同的应用程序中更节能和更具通用性

Robots with legged locomotion have, over the years, demonstrated high flexibility, excellent dynamic stability and adaptability to different terrains and obstacles [1]. They have demonstrated these characteristics in different environments, especially in areas such as rescue, reconnaissance, health-care, security and marine environments [2]. Robots have beenshown to be advantageous in applications involving dull, dirty and dangerous environments, and vital tasks such as nuclear decommissioning remain a key national priority. It is widely believed that over 50% of the earth's surface is inaccessible to wheels or tracks [2], and legged robots have an increasing role to play, judging by recent commercial successes such as ANYbotics [3] and cost-effective, open source systems such as Stanford Doggo [4] and the ODRI [5].Despite the development of legged robots demonstrating safety, robustness and high performance, it is evident that the commercially available robots are designed with centralised control, with every joint actuated. Here, it is pertinent to introduce Morphological Computation (MC), which, in the context of embodied (artificial) intelligence, refers to processes, which are conducted by the body (and environment) that otherwise would have to be performed by the brain [6]. MC is relevant in the study of biological and robotic systems as illustrated in the Passive Dynamic Walker [7], which is a purely mechanical system. The Passive Dynamic Walker shows that walking can result from the interaction of the system's physical properties and its environment without actuation. In the context of robotics, this means that systems with high morphological computation only need to generate motor commands when they are needed. Not only does such a control scheme increase the durability of the systems (because the wear-out of the actuators is decreased), it also means that robots with high MC will have a reduced energy demand for their actuation [8]. This is useful for legged robots, since they need to be untethered for autonomous functioning and transporting payloads over challenging terrains. While there is good acceptance of the role of MC in biological systems, Ghazi-Zahediet al [8], in their exploration on recenttrends, state that its application in robotics remains underexplored. One of the main reasons emphasised by the authors is that the conventional control paradigms treat the body as something that needs to be dominated rather than being used as a computing resource. There is a tendency to suppress any undesirable morphological behaviours like nonlinearity, underactuation or noise through the use of control systems and servo motors. It is quite remarkable that these same complex morphological properties play a key role in the behaviour of natural systems, as described in the research by Abad et al in the MC of a goat hoof in slip reduction [9] and the Puppy [7], an under-actuated robot. At the same time, one of the challenges set out by Deimal et al [10] is to ensure that the systems take advantage of the morphology ('good MC') while avoiding harmful body-environment interactions with respect to the desired functionality ('bad MC'). To achieve the right balance of MC and active control, a formal method is required to measure MC in the system. In their paper detailed with algorithms, Ghazi-Zahedi et al [6] have demonstrated two methods of measuring MC, one of which is to compare behaviour complexity with controller complexity, the former by information of world states and the latter by information of sensor states. The conclusion from review of current landscape of academia and industry is that there is a huge potential for robotics systems that use the inter-play of computational power, centralised control, and morphological features to be more energy efficient and versatile across different applications