NRI: EAGER: Teaching Aerial Robots to Perch Like a Bat via AI-Guided Design and Control

本研究将为设计具有蝙蝠外形和飞行能力的小型无人机奠定科学技术基础。这些类似蝙蝠的无人机将是非侵入性的，可以安全地在共享空间中运行，为人类提供态势感知。我们以蝙蝠为灵感的设计也将具有碰撞容忍度，以适应现实世界中混乱、难以进入的环境。由此产生的技术可以显著改善智慧城市的公共安全和车辆动态交通控制，以及与监测环境灾害相关的成本效益。最终，无人机能够在大型系统中提供计算、通信和传感能力，例如住宅建筑、街道、建筑区域和州立公园。这些能力将产生巨大的社会影响和经济效益。此外，作为这个项目的结果，新一代的科学家和工程师将在解决理论和实验交叉的多学科挑战方面得到培训。该项目将创建程序和工具，以培训劳动力的新技能，包括仿生机器人，机器学习和人工智能，以及非线性控制理论。本研究采用人工智能引导框架，研究蝙蝠的栖息（即倒立着陆）、零路径飞行和悬停等飞行动作。由于蝙蝠翅膀运动的突出方面可以在低维子空间中表示，我们将对来自真实动物的飞行数据应用自编码变分推理方法，以提取其底层飞行模型的低维但可解释的嵌入。我们还将使用高保真虚拟环境进行3D建模，综合设计和验证提取的低维潜在变量。这也将导致通过数据驱动程序更好地理解蝙蝠的感觉反馈机制。我们的研究目标将简化设计仿生空中协同机器人的工程程序，以模仿目标动物的飞行行为，因此直接降低理解闭环控制和仿生多模态协同机器人设计基础的障碍。为了实现提出的研究目标，我们将在这个项目的一年时间内集中精力进行两个主要阶段：第一，人工智能引导的蝙蝠各种飞行动作的分析和建模，第二，开发一种柔软和耐碰撞的蝙蝠启发的空中代理，能够降落在结构上。这两个阶段将通过团队的跨学科合作来完成，因为组件是高度相互联系和依赖的。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This research will develop the scientific and technological foundations for design of small unmanned aerial vehicles (UAVs) with bat shapes and flying abilities. These bat-like UAVs will be non-intrusive and safe to operate in shared spaces to provide situational awareness to humans. Our bat-inspired design will also be collision-tolerant to negotiate cluttered, hard-to-access environments in the physical world. The resulting technology can significantly improve public safety and vehicular dynamic traffic control in smart cities and cost-effectiveness associated with monitoring environmental disasters. Ultimately, the UAV can provide computing, communication and sensing capabilities in large-scale systems such as, residential buildings, streets, construction zones, and state parks. These capabilities should result in enormous societal impact and economic benefit. In addition, as the result of this project, a new generation of scientists and engineers will be trained in addressing multidisciplinary challenges at the intersection of theory and experiment. The project will create programs and tools to train workforce with new skills including bio-inspired robotics, machine learning and artificial intelligence, and nonlinear control theory. This research adopts an artificial intelligence-guided framework to study bat's flight maneuvers including perching (i.e. upside-down landing), zero-path flight, and hovering. Due to the fact that the salient aspects of the bat's wing motion can be represented in a low-dimensional subspace, we will apply an auto-encoding variational inference approach on the flying data from real animal in order to extract low dimensional, yet interpretable, embedding of the its underlying flight model. We will also use a high-fidelity virtual environment for 3D modeling, synthetic design and validation of the extracted low-dimensional latent variables. This will also lead to better understating of the bat sensory feedback mechanism through a data-driven procedure. Our research objective will simplify the engineering procedure to design bio-inspired aerial co-robots that closely mimic the flight behavior of a target animal, therefore is directly towards lowering the barriers for understanding fundamentals regarding closed-loop control and design of bio-inspired multimodal co-robots. In order to achieve the proposed research objectives, we will center our effort around conducting two main phases during the one year timeline of this project: first, AI-guided analysis and modeling of bat's various flight maneuvers, and second, development of a soft and collision-tolerant bat-inspired aerial agent capable of landing on structures. These two phases will be accomplished by team's cross-disciplinary collaborative, as components are highly interlinked and dependent.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.