ニューラルネットワークによる海洋ロボット用マニピュレータのエネルギー最小化軌道の実時間設定法

概要

論文の詳細を見る
Recently, ROV (Remotely Operated Vehicle) is often used in the development of the ocean resources. But the energy of the ocean robot is limited. So to put the untethered type ROV to practical use, we studied about trajectory planning method of the underwater manipulator according to the minimum control energy criterion. This problem is formulated as Two Point Boundary Value Problem (TPBVP), and it takes too much time to slove this problem numerically using Newton-like method. In this paper, we propose a simple trajectory planning method that generates minimum control energy trajectory in real time. This method approximates the solution of the TPBVP using fourth order polynomial and multi-layer neural network. The polynomial satisfy the boundary conditions at initial and final point of the trajectory, and pass through the characteristic point like apex of the trajectory. Neural network is trained by backpropagation algorithm in order to get the position of the characteristic point and the arrival time to the point when the position of initial and final point and final time required to move from the start to the end are given. In the case of the 2-joint underwater manipulator moving in a vertical plane, we examine the accuracy of the proposed method and the availability of the neural network. As a result, the necessary architechture of the neural network-number of the hidden layer, number of the units in each layer -to approximate the minimum control energy trajectory is presented. And it is proved that the trajectory planning method proposed in this report can accurately approximate the solution of TPBVP, and the error of the energy consumption is about from 5 to 10% compared with the solution of TPBVP. By the ability of the neural network, this method can generate the trajectory for the condition not learned.
社団法人日本船舶海洋工学会の論文
1994-09-04

ニューラルネットワークによる海洋ロボット用マニピュレータのエネルギー最小化軌道の実時間設定法

スポンサーリンク

概要

著者

関連論文

スポンサーリンク