Kali: a knowledge-based system for grasp planning

Automatic planning involves a key problem: what is the right gripper and the right grasp for a given piece? The problem is so complex and interrelated with the task planning, that only a full informed planner can choose gripper and grasp. Kali (Knowledge Aided Ledges Inspector) is a knowledge-based grasp advisor that proposes a set of possible grasps compatible with the hypothesis received by a planner and all physical, geometrical, and mechanical problem constraints: it does not look for the ´best´ grasp. Kali is based on a form feature extraction approach. For each form feature suitable for grasping Kali refines step by step the domains of possible grasps, taking care of even more constraints. Under large assumptions the domains of possible grasps for a given form feature of a given object, with a given gripper, in a given scene can be represented as solid volumes in a 3-D space. Inputs and outputs of Kali are ASCII file in formal description CAD languages. Kali is developed in Common Lisp and KEE on a Unix workstation.<>