Reactive Algorithms in Robotics

本研究的主要目标是从高层次任务的描述出发，构建一个理解和设计电抗器件的理论框架。 这项研究是基于一类新的机器人算法，PI称之为“反应算法”。“这些提供了关于机器人本身如何像“模拟计算机”一样使用其身体部位的见解，并且在计算上比其“数字大脑”更强大，”也许最终可能会揭示一些问题，即神经元比机器人少得多的昆虫如何能够完成更复杂的操作和运动任务。 到目前为止，这些想法的主要应用已经在建造新的平行颚夹持器和多指（2-和3-指）手，一个“旋转机”和步行机。 PI基于这些算法思想建立了一个原型平行钳口夹持器（“纽约大学反应式夹持器”）。 由此产生的夹持器将必要的计算能力减少到仅几个简单的数字电路，利用原始的感测能力，即，不超过12个红外发射器和检测器，并且以平稳的方式操作，而不会干扰或损坏被操纵的物体。 PI一直在研究如何描述这些设备，并使用Ramadge-Wonham离散事件系统（DES）形式主义证明其正确性。另一个目标是基于用于分析在线算法与理想化的透视算法相比的“竞争性”的框架来理解这些设备的“计算”复杂性。其他理论问题涉及噪声的影响，算法修改免疫噪声，传感器的位置问题，稳定性和收敛性。

The primary goal of this research is to construct a theoretical framework for understanding and design of reactive devices starting from the description of a high-level task. The research is based on a new class of robot algorithms that the PI calls ``reactive algorithms.'' These offer insights into how a robot itself can use its body parts like "analog computers'' and be computationally more powerful than its "digital brain,'' and perhaps may ultimately shed some light on the question of how insects with far fewer neurons than a robot are able to accomplish much more complicated manipulation and locomotion tasks. So far the main applications of these ideas have been in constructing new parallel-jaw grippers and multifingered (2- and 3-fingered) hands, a "twirling-machine'' and a walking machine. The PI has built a prototype parallel-jaw gripper ("NYU reactive gripper'') based on these algorithmic ideas. The resulting gripper reduces the necessary computing power to only a few simple digital circuits, utilizes primitive sensing abilities, i.e., no more than a dozen infra-red emitters and detectors and operates in a smooth manner without disturbing or damaging the manipulated objects. The PI has been investigating how to describe these devices and prove their correctness using the Ramadge-Wonham Discrete Event System (DES) formalisms. Another goal is to understand the "computational'' complexity of these devices based on a framework for analyzing the "competitiveness'' of on- line algorithms in comparison to an idealized clairvoyant algorithm. Other theoretical questions relate to the effect of noise, algorithmic modification for immunization against noise, the sensor-placement problem, and stability and convergence properties.