Vision-Based Real Time Trajectory Adjustment for Brachiation Robot

We propose a brachiation robot named Brachiator. Brachiator is able to swing like a real gibbon. The limbs of Brachiatior are designed to be actuated through the wires. However, the elongation and slack of the wires often arise and make it difficult to estimate the joint angles accurately, which has to be resolved to attain more stable brachiation behavior. In this paper we introduce a vision system to improve brachiation behavior in real time. A stereo vision system is mounted on Brachiator that looks like a gibbon's face. We estimate errors between a model based trajectory and a real trajectory in implementation world from the images which the vision system captures during the brachiation behavior. Based on the vision-based error estimation as to the grasping position on the target branch, the trajectory of the behavior is adjusted in real time in order to realize successful brachiation behaviors. Experimental results showed that the proposed system reduced deviation of the grasping position of brachiation behavior as much as 43% compared with previous methods.