Polyhedral convex cones for computing feasible grasping regions from vision

This paper presents a method to compute a feasible region for grasping unknown objects using visual features. The method is divided into two separate sections: candidates grasp generation and testing of the feasible region of grasping by constructing a set of constraints for each candidate grasps. From visual features, the algorithm examines at least one possible candidate grasp region around the object's boundary for each available finger and selects the optimal grasping points on it, relative to some criteria for optimality. An experimental environment was developed to test this method.