Grasp Planning for a Humanoid Hand

We focus on grasp planning for a humanoid multi-fingered hand attached at the tip of a humanoid robot’s arm. The hand has potential possibility to grasp various objects under several situations. Since the multi-fingered hand can use several grasp types such as fingertip grasp, and envelope grasp with taking the advantage of degrees of freedom. We develop grasp planner which selects a feasible grasp type based on the task, and determines contact positions for the fingers and the grasped object surface so that the fingers do not drop the object while staying with limited actuator capacity. To grasp an object, the robot first measures object position/orientation using vision sensor. Then, the planner plans the bodymotion to complete the grasping task based on vision sensor information. Even when the object’s location is not known beforehand, the robot should complete the grasping task as fast as possible. However, grasp planning with a humanoid robot is complex and often requires long calculation time. Hence, for the grasp planning, a heuristic but fast algorithm is preferred rather than precise but slow algorithms (Shimoga (1989)). Our planner calculates grasp motions within reasonable time by using predefined grasp types which are assigned with contacting finger links, desired sizes of the grasped object. Our planner selects a grasp type according to position/orientation of the grasped object similar to a human. As shown in Fig. 1 , a human grasps the side of the can with all fingers, grasps the top with fewer fingers. Failing to find feasible grasping posture using arm/hand kinematics alone, our planner attempts to do so using the full body kinematics. Using the degrees of freedom of full body, the planner has adaptable for reaching the object with the several motions such as twisting waist, bending waist, and squatting down. We demonstrate effectiveness of grasp planning through simulation and experimental results by using humanoid robot HRP-3P (Akachi et al. (2005)) shown in Fig. 2, which has a four-fingered hand HDH (Kaneko et al. (2007)) on its right-arm and a stereo camera system in its head. We has proposed the grasp planning which was executed within the reasonable time taking into account several constraints imposed on the system such as the feasible grasping style, the friction cone constraint at each contact point, the maximum contact force applied by the fingers, the inverse kinematics of the arm and the fingers (Harada et al. (2008); Tsuji et al. (2008; 2009)). The originality of our method is the following points; 4