Autonomous walkers and explorers: collective dynamics of coupled oscillators for autonomous robot locomotion and coordination

通过将几个高风险的任务委托给自治系统，自动化方面的进步对于提高生活质量是至关重要的。例如，机器人助手、有毒环境中的维护和监测，以及受污染生态系统的质量控制。此外，机器人系统可以与生物系统相结合，在人群动力学和控制等特定场景中共同驱动它们，例如在突发危险造成的疏散情况下。在设计一类专门用于这些操作的自主系统时，一个决定性的基本特征是能够可靠地协调不断变化的环境，并且相对于名义操作条件具有高度的不确定性。这对设计移动系统提出了一系列挑战，因为机械效率方面的典型约束通过高度专业化得到最佳满足，而健壮性和通用性则通过低专业化和在不同环境中最佳平均行为方面的权衡很好地解决。通过纤毛和鞭毛等毛状结构的发育，小规模的生物结构通过进化过程获得了强大的运动特征。最近的研究揭示了这些结构的基本性质和普遍性，发现了更多的功能，如微观尺度上的射流传输，以及更复杂的功能，如管理异质生物系统中的共生相互作用，以及调节胚胎发育过程。这类系统协调动力学的高度通用性和健壮性使其对包括低专业化和多功能的运动机器人在内的应用具有吸引力。因此，至关重要的是研究耦合的基本物理，它可以诱导以适合运动的波传播现象的形式出现协调和集体动力学，从而为设计具有多种特征的健壮自主步行器提供信息。对这类耦合振子系统中新出现的协调现象的非线性动力学的研究扩展到自主机器人网络，即从行走的协调(个体)到任务导向的任务中的群体协调，例如，在监测脆弱环境和随后的有针对性的干预方面。基于中心模式生成器模型的机器学习已被应用于机器人学，以模拟动物运动背后的基本神经结构。深度学习和集体运动的基本非线性动力学之间的联系将被研究，以获得关于自然界中出现集体模式的必要性的见解，以及关于适合复制它们的神经网络结构的见解。

Advances in automation are fundamental in contributing to the improvement of the quality of life, by delegating to autonomous systems several high-risk tasks. Examples involve robotic assistants, maintenance and monitoring in toxic environments, and quality control of contaminated ecosystems. Additionally, robotic systems can be coupled with biological systems, to co-drive them in specific scenarios such as crowd dynamics and control, for example in evacuation situations caused by sudden hazards. In designing a class of autonomous systems dedicated to these operations, a defining fundamental characteristic is the capability of reliably negotiating an evolving environment, with high degrees of uncertainty with respect to nominal operating conditions. This poses a set of challenges in designing a locomotion system, since typical constraints in terms of mechanical efficiency are optimally fulfilled through high-specialization, whereas robustness and versatility are well addressed through low-specialization and trade-offs in terms of best averaged behavior across different environments. Robust locomotion features have been acquired through evolutionary processes by small scale biological structures through the development of hair-like structures such as cilia and flagella. Recent research revealed the fundamental nature and ubiquity of these structures, with the discovery of additional functions such as fluidic transport at the microscale, as well as more complex functions such as managing of symbiotic interactions in heterogeneous biosystems, and the regulation of processes in embryonic development. The high versatility and robustness of the coordinated dynamics of this class of systems make it attractive for applications that include locomotion robotics characterized by low-specialization and multi-function. It is therefore crucial to study the fundamental physics of the coupling that can induce the emergence of coordination and collective dynamics in the form of wave propagating phenomena that are suitable for locomotion, to inform the design of robust autonomous walkers with versatile characteristics.  The study of the nonlinear dynamics of emerging coordination phenomena in this class of systems of coupled oscillators extends to networks of autonomous robots, that is, from coordination for walking (individual) to group coordination in task-oriented missions that may be broadly defined, for example in terms of monitoring vulnerable environments and subsequent targeted interventions. Machine learning based on central pattern generator models has been applied to robotics to mimic the fundamental neural structures behind animal locomotion. The connection between deep learning and the underlying nonlinear dynamics of collective motions will be investigated to gain insights about the necessity of collective patterns emerging in nature, and about neural networks architectures suitable to reproduce them.