Collaborative Research: FRR: Adaptive mechanics, learning and intelligent control improve soft robotic grasping

处理柔软、易碎或光滑的物体，如成熟的水果，在机器人技术中仍然是一个挑战。软式机器人抓取器在安全处理此类物体而不损坏它们方面显示出巨大的希望。此外，开发控制软机器人的软件还带来了另一个挑战。相比之下，许多身体柔软的动物在觅食和进食时每天都会解决这个问题。它们不仅能够抓取和操纵柔软易碎的物体，而且动物还可以学习如何安全地与新物体互动，并根据先前的经验改变它们在抓取过程中施加的力。该项目将从一种有身体的动物--海懒身上获得灵感，这种动物成功地以一系列大小、韧性和形状差异很大的海藻为食，创造出一种新型的软抓取机器人。该项目还将创建一种机制，可以学习如何安全地抓住各种物体，包括西红柿和蘑菇等易碎食物。机器人学习如何安全地处理柔软和脆弱的物体的能力将在未来的农业、制造业和医学中得到应用。该项目还将支持对科学和工程领域多样化劳动力的培训。从小学到大学的学生将被作为研究参与者来测试机器人。此外，该项目将通过研究生培训、外展活动、本科生暑期研究经验和科学插图实习来支持跨学科培训。该项目将测试以下假设：具有板载生物启发学习和本地控制的软、形态智能抓取机器人将通过快速调整控制器和执行器的特性以及实时学习来提高抓取性能和易用性。为了验证这一假说，该项目将：(1)实现执行器在短时间尺度上的适应性，模仿生物肌肉的短期变化；(2)通过在合成神经系统(SNS)中的短期学习，模拟生物神经系统中的短期网络变化，实现局部控制适应性；以及(3)在SNS中实施长期突触权重变化，模仿从经验中学习。在目标1和目标2中，将采用一种生物启发的方法来开发一种软抓取工具，其灵感来自于海鞘(海懒)的摄食。在目标3中，这种方法将扩展到机器人手臂，并将长期学习纳入控制器。为了准确识别需要学习的网络元素，这个项目将研究易驯服的动物模型--加州海兔的抓取能力。这种海洋海懒擅长抓取柔软、易碎、易滑的物体，并凭借经验快速学习。此外，海兔的抓握控制电路只包含几百个神经元，可以测量学习过程中关键网络元素的具体变化。为了评估抓取的生物学原理的价值，该项目将使用人类受试者来衡量机器人抓取器的性能、易用性和操作员培训时间。该项目由机器人基础研究项目支持，由工程指导委员会(ENG)和计算机与信息科学与工程指导委员会(CEISE)共同管理和资助。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Handling soft, fragile, or slippery objects such as ripe fruit remains a challenge in robotics. Soft robotic graspers show tremendous promise in safely handling such objects without damaging them. Furthermore, creating software to control soft robots poses an additional challenge. In contrast, many animals with soft bodies solve this problem everyday as they forage and feed. Not only are they able to grasp and manipulate soft and fragile objects, but animals can also learn how to safely interact with new objects and vary how much force they apply during grasping based on their prior experience. This project will take inspiration from an animal with a body, the sea slug, that feeds successfully on a range of seaweeds that vary greatly in size, toughness and shape, to create new type of soft grasping robot. This project will also create a mechanism that can learn how to safely grasp a wide range of objects, including fragile foods like tomatoes and mushrooms. The ability for a robot to learn how to safely handle soft and fragile objects will have future applications in agriculture, manufacturing, and medicine. This project will also support the training of a diverse workforce in science and engineering. Students from grade school through college will be included as research participants to test the robot. Additionally, this project will support cross-disciplinary training through graduate student training, outreach activities, summer research experiences for undergraduates, and internships in scientific illustration.This project will test the hypothesis that soft, morphologically intelligent grasping robots with onboard bioinspired learning and local control will improve grasping performance and ease of use by rapidly adjusting controller and actuator properties and learning in real-time. To test this hypothesis, this project will: (1) implement actuator adaptability over short timescales, mimicking short-term changes in biological muscle, (2) implement local control adaptability through short-term learning in a synthetic nervous system (SNS), mimicking short-term network changes in biological neural systems, and (3) implement longer-term synaptic weight changes in an SNS, mimicking learning from experience. In Aims 1 and 2, a bioinspired approach will be applied to develop a soft grasper inspired by Aplysia californica (sea slug) feeding. In Aim 3, this approach will be extended to a robot arm and long-term learning will be incorporated into the controller. To precisely identify elements of the network subject to learning, this project will study grasping in a tractable animal model, Aplysia californica. This marine sea slug is adept at grasping soft, fragile, slippery objects and rapidly learns with experience. Furthermore, Aplysia’s grasping control circuitry contains only a few hundred neurons, allowing the measurement of specific changes in key network elements during learning. To assess the value of biological principles for grasping, this project will use human subjects to measure the robotic grasper’s performance, ease of use, and operator training time. Baseline data will be established with a conventional grasper and performance will be compared as adaptability is integrated into the system.This project is supported by the cross-directorate Foundational Research in Robotics program, jointly managed and funded by the Directorates for Engineering (ENG) and Computer and Information Science and Engineering (CISE).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.