PFI-TT: Robotic Dexterity for Material Handling

创新伙伴关系-技术转化（PFI-TT）项目的广泛影响/商业潜力包括为我们的经济建立更有效和更强大的供应链的步骤。近年来，物料搬运行业对我国国民经济的重要性日益凸显：混乱的供应链在社会中无处不在。该行业面临的最大挑战是人体工程学难题和重复性使得许多任务容易受伤，不适合工人。然而，自动化只在很小程度上缓解了这个问题：机器人在物料搬运方面只取得了很小的市场渗透，其中一个关键原因是缺乏与多功能和灵巧的机器人物料搬运相关的关键技术。目前，该领域还缺乏能够处理灵巧任务的柔性机器人。开发这样的机械手将增强我们对运动控制任务和运动学习的科学理解，并对提高供应链效率产生直接影响。拟议中的项目旨在开发这种技术并将其转化为市场。我们以首席研究员团队最近的工作为基础，该团队展示了复杂的机器人操作任务，例如用手指步态重新定位大型手持物体，同时保护被操作物体。这一结果是通过将运动技能深度强化学习的新探索方法与PFI团队先前开发的基于光学的触觉手指相结合而实现的。该项目旨在进一步开发基于学习的外部操作方法，其中机器人使用外部表面来传递所需的运动，这一策略将有助于减少对手部本身的运动学要求。这将反过来使人手能够安装在商用机器人手臂上，从而形成一个完整的自动化站。虽然之前的工作仅依赖于触觉和本体感觉来展示PFI项目的灵活性，但完成任务的部署需要添加视觉传感器，因此需要学习多模态、视觉控制策略的方法来实现灵巧的外部操作。PFI团队将把这项研究应用于自动化分拣系统感应过程的具体任务，其中包括灵活的包装重新定位，这是配送中心作为供应链支柱的一项常见任务。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 broader impact/commercial potential of this Partnerships for Innovation - Technology Translation (PFI-TT) project consists of steps towards a more effective and robust supply chain for our economy. Recent years have highlighted the critical importance of the material handling industry to our national economy: a snarled supply chain reverberates everywhere in society. The number one challenge in this industry is that the difficult ergonomics and repetitive nature make many tasks injury-prone and unsuitable for workers. Nevertheless, automation has alleviated this problem only to a small degree: robotics has achieved only small market penetration in material handling, and one of the key reasons is the absence of critical technology pertaining to versatile and dexterous robotic material handling. Currently, flexible robotic manipulators able to handle dexterous tasks are largely lacking from the field. Developing such manipulators would both enhance our scientific understanding of motor control tasks and motor learning and have an immediate impact in increasing supply chain efficiency.The proposed project aims to develop and translate such technology to the marketplace. We build on recent work from the principal investigator’s team, which demonstrated complex robotic manipulation tasks, such as large in-hand object reorientation with finger gaiting, while simultaneously securing the manipulated object. This result was achieved by combining novel exploration methods for deep reinforcement learning of motor skills with the optics-based tactile fingers previously developed by the PFI team. The project aims to further develop learning-based methods for extrinsic manipulation, where the robot uses external surfaces to impart the desired movement, a strategy that will help reduce the kinematic requirements on the hand itself. This will in turn enable hands to be mounted on commercial robot arms, resulting in a complete automation station. While the previous work informing the PFI project relied exclusively on tactile and proprioceptive sensing to demonstrate dexterity, deployment for complete tasks requires the addition of vision sensors, and thus methods for learning multimodal, visuotactile control policies for dexterous extrinsic manipulation. The PFI team will apply this research to the concrete task of automating the sortation system induction process with dexterous package re-orientation, a commonly found task in the distribution centers that are the backbone of our supply chain.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.