3D Camera Identification for Enabling Robotic Manipulation

Many autonomous robotic systems have a need to sense the world around them in order to safely interact with and manipulate objects while avoiding unintentional collisions with other entities. For systems that require fairly precise localization of objects over long ranges, the data needed to perform such actions has historically been collected through expensive laser based devices such as LADARS. For less precise applications, researchers have utilized standard 2D cameras (with and without tags), 3D stereoscopic camera systems, RADAR systems (such as those used in many automotive driver assist systems), and ultrasonic sensors. Recently, a new type of actively illuminated 3D camera has become available that can provide high resolution and relatively large ranges (although not as long as LADARs) but at a price that is more comparable with the less expensive, low resolution sensors. While this hardware provides valuable data about the world, it requires new techniques for processing the data to enable intelligent interpretation by the robotic systems. To this end, in this presentation we detail the unique qualities of this new sensor system and the techniques required for processing its data to support object manipulation tasks such as: Autonomous engagement of an object resting on the floor, resting on a stack of similar objects, or resting on a platform (including larger objects) Autonomous placement of an object on the floor, a stack of similar objects, or on platform, Detection and avoidance of unintentional collisions with objects, We present the methods used, results of our work, current challenges, and future approaches we are taking with our research.