This paper presents a randomized manipulation planner for a multi-fingered hand by switching contact modes. Manipulation planning for such a system should consider changing kinematics and dynamics of the system according to the contact modes. We derive the conditions that would be satisfied in manipulation planning, based on the kinematical properties of a manipulation system. The conditions allow the planner to assign more contact modes than those corresponding to the relative motion of contact points restricted by constraint forces. Consequently, this enables dextrous manipulation. In addition, the conditions give the restrictions of feasible contact modes and the number of contact points. Inspired by randomized motion planning techniques, we propose a new algorithm for manipulation planning in order to explore object configuration space, SE(3), rapidly and uniformly. The basis for this approach is for the construction of exploring random trees. Simulation examples for 3-D manipulation by switching contact modes are presented to verify the planner's effectiveness.

We propose a novel complementarity model for a general three-dimensional manipulation system with rolling and sliding contacts to solve the forward dynamics problem. The key idea is to derive complementary constraints for a rolling contact by decomposing the tangential properties. We also discuss the existence and uniqueness of a solution to the model, and algorithms for solving the problem. We show examples of computer simulation for dynamics of multifingered hand manipulation and discuss how the transition of contact modes varies for applied joint driving torques.

We present motion planning for dexterous manipulation by a whole arm grasp system based on switching contact modes. Motion planning for such a system should consider changing dynamics and kinematics according to the contact modes. We systematize the properties of the manipulation system by taking account of contact dynamics and kinematics, and derive the conditions to be satisfied in motion planning. The conditions give the restrictions of the feasible contact mode and the number of contact points. We propose an algorithm of global motion planning based on a hybrid control perspective. We aim at finding a feasible joint torque trajectory and a sequence of contact modes to move the object to an object final state for given switching times and object state trajectory. The proposed planning algorithm mainly consists of planning the object trajectory by a randomized approach using an object nominal trajectory, and constructing subgoals which satisfy the proposed conditions by solving an inverse problem. We show the validity of the planner by simulating manipulation by whole arm grasp system.