Robotic exploration of material and kinematic properties of objects

摘要

The physical interaction with unstructured environments requires that robotic systems have the ability to extract material and kinematic properties of objects around them. The goal of this research is to design a robotic system that actively explores and extracts the material properties, including thermal, hardness and mass properties and of kinematic properties, such as mobility and geometric parameters of objects and their parts. To accomplish this objective, we invoke the paradigms of active perception and exploratory procedures. We develop methodologies for the design of such procedures a well as sensors which support their use in the robotic domain and demonstrate their effectiveness. The system is composed of a control module which coordinates the visual and the haptic sub-modules. Vision is implemented via an agile laser range-scanner which is able to acquire different views of the desired object. Global volumetric models of the object are recovered by fitting super-ellipsoids to the 2${1\over 2}$ D range image. The haptic module uses the geometric information of the object to perform several tests based on non-destructive techniques. For exploring thermal properties, a new approach for the design and modeling of thermal sensors for robotics is presented. A model of this sensor is developed and its validity is experimentally verified with different materials. Mass density is estimated by the weight evaluation procedure. Hardness is evaluated by means of stress vs. strain tests. Compression and tension tests are performed to determine this property. The kinematic characteristics of the object are explored by the mobility procedure. We describe a novel methodology, based on screw theoretic results which enables the identification of the mobility of the object. This is accomplished by forming a closed kinematic chain with the manipulator and the unknown object. The number of degrees of freedom present in the object as well as the geometric parameters of its links are then extracted. The design and implementation of the robotic haptic architecture testbed where all of the above concepts were smoothly integrated into a working system is also described. The architecture controls and coordinates the two robot manipulators, the instrumented parallel-jaw gripper and the mobile laser range-scanner.