Dynamic object localization via a proximity sensor network

Abstract

In this paper we describe a proximity sensor network (PSN) consisting solely of inexpensive intensity-based electro-optical proximity sensors embedded in a robotic end-effector. By coupling the PSN to an extended Kalman filter, and by resolving the sensors' dependence on ambient light, the object's reflective properties and the angle between sensor beam and the object's surface, we succeed in estimating a cylindrical moving object's unknown planar trajectory. Experiments show robust position and velocity estimation of the moving object, despite noisy sensor data and unmodelled object dynamics. Such a system should be directly applicable to sensor-based control approaches for dynamic grasping and manipulation in robotics and automation.