NRI: FND: Foundations for Physical Co-Manipulation with Mixed Teams of Humans and Soft Robots

这个国家机器人计划（NRI）项目的目标是使人类和机器人的混合团队能够一起工作，在复杂的环境中完成对物理要求很高的物体操作任务。对于这个项目，软机器人是专门考虑的，因为传统的机器人太重了，而且有潜在的危险，无法与人密切合作。人类能够有效地合作移动一个笨重的物体，因为他们能够利用他们对群体目标的理解和个人能力来解释物理线索，并迅速推断出彼此的意图。因此，将这种能力扩展到机器人的第一步是了解一群人如何识别和反应来自其他团队成员的推拉。该项目还强调了在与软机器人和人一起工作时管理不确定性的必要性——软机器人是因为它们在典型的任务负载下会显著变形，而人是因为机器人可能难以预测他们的动作。该研究的潜在应用范围可以从加快物流和材料处理，到改善采矿、石油钻井平台、伐木和搜救等危险和/或难以到达的环境中的人类安全。为此目的，将与当地的搜救队合作，在整个项目中征求对人机协同操作的反馈。代表性不足的本科生将接受利用软机器人技术的STEM教育工具的培训，然后学生们将努力将这种培训传播到当地的K-12教室。协同操作可以定义为许多协作代理在移动单个大型对象时所采取的行动和发送的信号。本研究将实现人与机器人之间的协同操作，重点关注以下三个主要方向：1)建模、控制和规划软机器人处理任务不确定性的有效刚度轨迹；2)在操作过程中量化和建模人类意图和共识；3)开发包含意图、共识和不确定性的算法来执行协同操作任务。基于先前对大自由度软机器人模型预测控制算法的研究，基于任务不确定性的估计，将生成刚度轨迹作为软机器人控制的一部分。人类合作者在现实生活和虚拟现实中移动大型物体的试验，将使预测群体共识和运动的算法得以发展。最后，在给定群体短期运动目标的合理估计的情况下，生成的算法还将生成机器人运动和刚度轨迹，从而通过减少不确定性来帮助群体更有效地达成共识。这项研究将开创自然物理交互、安全机器人控制、多智能体协调和以分布式方式规划/行动的新组合。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。

The goal of this National Robotics Initiative (NRI) project is to enable mixed teams of humans and robots to work together to accomplish physically demanding object manipulation tasks in complex environments. For this project, soft robots are exclusively considered, because traditional robots are too heavy and potentially dangerous to work closely with people. Humans can effectively work together to move a bulky, heavy object because they are able to use their understanding of group goals and individual capabilities to interpret physical cues and quickly infer each other's intention. Thus, the first step in extending this ability to robots is to understand how groups of people recognize and react to pushing and pulling from other team members. The project also emphasizes the necessity of managing uncertainty when working with soft robots and with people -- soft robots because they deform significantly under typical task loads, and people because their movements may be difficult for robots to predict. Potential applications of the research can range from expediting logistics and material handling, to improving human safety in dangerous and/or hard-to-reach environments such as mining, oil rigs, logging, and search and rescue. To this end, a collaboration with a local search and rescue team will solicit feedback on human-robot co-manipulation throughout the project. Underrepresented undergraduate students will be trained with a STEM education tool leveraging soft robotics, and the students will then work to disseminate this training to local K-12 classrooms. Co-manipulation can be defined as the actions taken and the signals sent by many collaborating agents while moving a single large object. This research will enable co-manipulation between humans and robots, and is focused on the following three main thrusts: 1) modeling, controlling, and planning effective stiffness trajectories for soft robots to deal with task uncertainty, 2) quantifying and modeling human intention and consensus during manipulation, and 3) developing algorithms that incorporate intention, consensus, and uncertainty to execute co-manipulation tasks. Building on prior work on model predictive control algorithms for large-degree-of-freedom soft robots, stiffness trajectories will be generated as part of the soft robot control based on estimates of task uncertainty. Trials with human collaborators moving large objects in real life and in virtual reality will allow the development of algorithms that predict consensus and motion of the group. Finally, given a reasonable estimate of the short-term motion goal of a group, the resulting algorithms will also generate robot motion and stiffness trajectories to help a group reach consensus more efficiently by reducing uncertainty. This research will pioneer the novel combination of natural physical interaction, control for safe robots, multi-agent coordination, and planning/acting in a distributed manner.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.