Active Vision Based Regrasp Planning for Capture of a Deforming Object Using Genetic Algorithms

Abstract

The ability to efficiently grasp an object is the basic need of any robotic system. This research aims to develop an active vision based regrasp planning algorithm for grasping a deforming 2D prismatic object using genetic algorithms (GA). The possible applications of the proposed method are in areas of grasping biological tissues or organs, partially occluded objects and objects whose boundaries change slowly. Most previous studies on robotic grasping mainly deal with formulating the necessary conditions for testing grasp points for static objects (Blake (1995), Chinellato et al. (2003), Galta et al. (2004), Mirtich et al. (1994)). Nguyen (1989) has suggested a strategy for constructing an optimal grasp using finger stiffness grasp potentials. A detailed review of multifinger grasping of rigid objects is presented in Bichi and Kumar (2000). There are few studies on grasping of deformable objects, such as Hirai et al. (2001) in which they present a control strategy for grasping and manipulation of a deformable object using a vision system. In this case the object deforms on application of fingertip forces, the deformation is recorded by a vision systems and based on the amount of deformation the object motion is controlled. Studies relating to searching and tracking of grasping configurations for deforming object are rare. Deforming objects are those that deform by themselves without application of external forces. Mishra et al. (2006) have proposed a method of finding the optimal grasp points for a slowly deforming object using a population based stochastic search strategy. Using this method it is possible to find the optimal grasp points satisfying force closure for 2D prismatic deforming objects. This method minimizes the distance between the intersection of fingertip normals and the object centre of gravity, and maximizes the area formed by the finger tip contact points. However their method fails in cases when the fingertip normals do not intersect at a point (as in case of a square object). The problem of grasping deforming objects is a very challenging problem as the object shape changes with deformation. Hence the optimal grasp points have to be continuously found for each new shape. This process of recalculating the fingertip grasp points due to object shape change, slide or roll is called regrasping. The best method of determining the change in shape of an object is by using a vision system. A vision system not only captures the new shape but can also be used to track a moving object. The main objectives of this research are to use a vision system to capture the shape of a deforming object, divide the