NRI: FND: Bioinspired Design and Shared Autonomy for Underwater Robots with Soft Limbs

目前，复杂而具有挑战性的水下操作任务只有在人类潜水员能够安全进入环境的情况下才能完成。机器人可以帮助解决这个问题，使人类操作员远离危险环境，同时仍然利用他们的技能来规划复杂的操作任务。然而，刚性机械臂通常不适合精细的操作任务。为了解决这一限制，该项目探索了受章鱼启发的软机械臂的设计和控制。软机器人可以在水下环境中执行复杂的科学任务，比如提取生物样本或精致的人工制品，而不会对它们造成伤害。这种软体机器人在与人类一起工作时也会更安全。研究结果将作为俄勒冈州立大学机器人博士课程的一部分整合到实践课程中，并通过俄勒冈州立大学机器人俱乐部和俄勒冈第一机器人比赛纳入K-12推广。本研究的目的是建立一个水下操作框架，将人类操作员和机器人之间的共享自主权与机械定向软驱动和传感相结合。提出的工作将研究新的致动器形态，替代制造技术，以及可拉伸集成液态金属传感器的使用。为了控制软抓取器，该项目开发了一个规划和控制界面，该界面利用机器学习技术来利用人类操作员的技能快速识别稳定的抓取。软抓取器的物理属性将与算法一起设计，这将提高对水下交互和共享自主性的理解。对大接触面积、多点夹持的柔性机械臂的灵巧性和顺应性进行了评价，该方法尤其有利于水下精细物体的抓取。测试将在台式水下试验台进行，使用运动学运动捕捉和相互作用力来评估性能。

Currently, delicate and challenging underwater manipulation tasks are performed only by human divers when it is safe for them to enter the environment. Robots can help address this issue by distancing human operators from dangerous environments, while still leveraging their skills in planning complex manipulation tasks. Rigid robotic manipulators, however, are typically not suitable for delicate manipulation tasks. To address this limitation, this project explores the design and control of soft robotic arms inspired by the octopus. Soft robots can perform delicate tasks of scientific interest in underwater environments, such as retrieving biological samples or delicate artifacts, without harming them. Such soft robots can also be safer when operating alongside humans. The results of the research will be integrated into hands-on courses as part of the Oregon State University PhD in Robotics and into K-12 outreach via the OSU Robotics Club and Oregon FIRST Robotics competitions.The objective of this research is to establish a framework for underwater manipulation, combining shared autonomy between human operators and robots with mechanically-directed soft actuation and sensing. The proposed work will examine new actuator morphologies, alternative fabrications techniques, and the use of stretchable integrated liquid metal sensors. To control the soft grippers, this project develops a planning and control interface that utilizes machine learning techniques to leverage human operators' skills at quickly identifying stable grasps. The physical attributes of the soft grippers will be designed in tandem with algorithms, which will provide improved understanding of underwater interaction and shared autonomy. Dexterity and compliance of the soft manipulators will be evaluated for large contact-area, multi-point gripping, which is particularly advantageous for grasping delicate objects underwater. Testing will be done in a benchtop underwater test bed, using kinematic motion capture and interaction forces to evaluate performance.