Task-Level Perception From Kinesthetic Sensors

本研究探讨了移动操作任务中运动知觉数据对任务水平信息的知觉。移动操纵器通过其脚或轮子与地形交互，并通过其手、轮子、保险杠或其他效应器与固定或移动对象交互。所有这些相互作用都有助于机器人的运动和机器人的内部形状变形，这些变形可以通过编码器、应变仪和其他传感器检测到。该动觉数据可用于估计地形的特征、物体的位置、机器人的位置、机器人与物体之间的接触状态以及其他任务变量。该项目正在开发任务级运动知觉理论，通过开发合适的任务和机器人动力学模型，并使用非线性可观测性框架来分析特定任务，并开发估计任务参数和状态变量的算法。实验工作包括对几种移动机械手设计的评估，观察者的实现和测试，以及移动操作任务的演示。

This research explores perception of task-level information from kinesthetic sense data in mobile manipulation tasks. A mobile manipulator interacts with the terrain through its feet or wheels, and it interacts with fixed or mobile objects through its hands, wheels, bumpers, or other effectors. All of these interactions contribute to the robot's motion and to internal shape deformations of the robot, which can be detected by encoders, strain gauges, and other sensors. This kinesthetic sense data can be used to estimate features of the terrain, locations of objects, location of the robot, contact state between the robot and objects, and other task variables. This project is developing a theory of task-level kinesthetic perception, by developing suitable models of task and robot dynamics, and using the framework of nonlinear observability to analyze specific tasks and develop algorithms for estimating task parameters and state variables. Experimental work includes evaluation of several mobile manipulator designs, implementation and testing of observers, and demonstration in mobile manipulation tasks.