Collision Detection and Control of Parallel-Structured Flexible Manipulators Based on Unscented Kalman Filter

Abstract

Flexible manipulators that are lightweight and mechanically flexible are useful for operations in various fields, e.g. space development programs, robotic assistants to humans and so forth. However, the derivation of their exact mathematical model and synthesis of the accurate positioning controller is exceptionally difficult because of the mechanical flexibility. On the other hand, the mechanical flexibility is conducive to the safety in collision between the manipulator and the obstacles. However, it is not positive safety measures that reduction of the influence due to the impact force of collision for the flexible manipulator depends only on its mechanical flexibility. In order to develop flexible manipulators so that they work safely with persons cooperatively, we need to introduce the active collision detection and suspension control algorithms to the flexible manipulators. The functional requirements of safety in the operations of flexible manipulators are as follows: i) to avoid collisions with obstacles placed or moving in the work space; ii) to detect collisions when unlooked-for obstacles contact with the flexible arm of the manipulator and to suspend the motion as immediately as possible; iii) to plan a new path so as to avoid the place of obstacles. There are several researches on collision detection methods without extra sensors (A. Garcia & Somolinos, 2003), (M. Kaneko & Tsuji, 1998), (T. Matsumoto & Kosuge, 2000). Moorehead and Wang proposed (Moorehead & Wang, 1996) a collision detection method using strain gauges to determine the intensity and position of external force due to collision with a flexible cantilevered beam. The estimation of the contact position in their approach was achieved by the mechanical relation between positions of the two strain gauges and the bending moments measured by the sensors. Payo et al. (I. Payo & Cortazar, 2009) is produced the method of collision detection and suspend control of the very lightweight single-link flexible arm based on coupling torque feedback. They used the variation of the control torque. The authors have focused our attention on the second item mentioned above. We already developed a method of collision detection for the single-link flexible manipulator using the innovation process of the Kalman filter (Sawada, 2002a), (Sawada, 2002b), (Sawada, 2002c), (Sawada, 2004 (in Japanese)), (Kondo & Sawada, 2008). Our approach requires no particular