Bundle Trust Region法による最適把持位置計画

抄録

Grasp planning is a fundamental but difficult problem for multi-contacted grasp and dexterous manipulation. The problem is in general a constrained optimization with nondifferentiable complexity. Gradient or subgradient method is traditionally used for finding solution, however, it is much inefficient and time consuming. This paper presents a new application of bundle trust region method for computing the optimal hand grasp. The algorithm is computationally efficient in the sense of fast convergence and short computation time. The solution corresponds to a force-closured grasp on arbitrary shaped 3-D object with any number of contacts, and furthermore it maximizes the grasp ability to reject disturbance forces. Numerical tests validate the effectiveness of the proposal.