Human-Like Dexterous Manipulation using Compliant Actuations

机器人已经成为现代工业自动化的一个组成部分，很难想象没有机器人的自动化计划。然而，机器人的应用仅限于一个非常狭窄的范围，即明确定义和预先计划的任务。事实上，当涉及到操纵普通物体时，机器人达不到人们的期望。相比之下，人类有非凡的操纵物体的能力。它们通过长期的适应过程获得这种能力；2岁学会抓东西，5岁学会用勺子，10岁学会用牙刷。这个过程使人类能够协调手臂和手的运动作为一个整体。此外，人类的运动控制系统是非常宽容的，允许改变手的刚度（对外部干扰的反应力）迅速满足任务的要求。这项研究的长期愿景是为物体操纵开发新的创新解决方案，使机器人能够像人类一样灵活地执行任务。通过解决对象操作的各个方面的一系列短期努力，实现了这一愿景。第一个短期目标是开发一个统一控制机械臂和机械手的通用框架（抓取问题）作为一个组合优化问题。与人类不同，在机器人中，这两个问题传统上被视为两个独立的问题，具有次优且通常不可行的解决方案。第二个短期目标是利用柔性驱动器（一种使用非刚性元件与负载耦合的新型驱动器）为机器人提供类似于人类肌肉驱动器调节刚度的能力。执行器的调制符合最小的传感需求和计算时间，使机器人具有与生物手相似的性能，同时提供工业应用所需的强度和耐用性。第三个短期目标侧重于使用新形式的“智能”材料和新控制方法进一步提高柔性驱动的技术水平。

Robots have become an integral part of modern industrial automation and it is hard to imagine an automated plan without a robot. Yet, the application of robots is limited to a very narrow scope of well-defined and pre-planned tasks. In fact, when it comes to manipulating ordinary objects, robots fall short of the expectations. In contrast, humans have exceptional ability to manipulate objects. They acquire this ability through a long process of adaptation; 2 years to grasp things, 5 years to use a spoon, and 10 years to use a toothbrush. This process allows humans to coordinate the movements of the arm and the hand as one unit. In addition, human’s motor-control system is extremely forgiving allowing to alter the stiffness of the hand (the reaction forces to external disturbances) swiftly to match a task’s requirements. The long term vision of this research is to develop new and innovative solutions for object manipulation to enable robots perform tasks with human-like dexterity. This vision is enabled through a series of short-term efforts to address various aspects of object manipulation. The first short-term goal is to develop a general framework for unified control of the robot arm and the robot hand (grasp problem) as a combined optimization problem. Unlike humans, in robotics, these two problems are traditionally treated as two separate problems with sub-optimal and often unfeasible solutions. The second short-term goal is to exploit compliant actuations (a new class of actuators that couples to a load using a non-rigid element) to provide robots with similar capabilities as human muscle actuations for modulating stiffness. The modulation of the actuator compliance with minimal sensing needs and computation time enables a robot to have similar performance as a biological hand while providing the strength and durability required in industrial applications. The third short-term goal focuses on further enhancing the state of the art in compliant actuation using new forms of “smart” materials and new control approaches.