Optimal Sensor Placement for Obstacle Detection of Manipulator Based on Relative Entropy

An optimal sensor placement method based on relative entropy is presented with the aim of maximizing obstacle information with limited number of sensors for manipulator. Two distribution - sensor detection distribution (SDD) and obstacle appearance distribution (OAD) - are defined, and the relative entropy of SDD from OAD is taken as optimization objective. Then, SDD is calculated by Monte Carlo simulation with the assist of collision detection between sensors and environment. A numerical computation method for the optimal placement is proposed in order to dealing with the problem of non-analytic expression of SDD. Performances of best and worst placement are compared though numerical example. The result shows that the best placement leads to a mission success rate about 29% higher than the worst.