Controle cinemático de robôs cooperativos usados na manipulação de objetos.

Abstract

This work addresses a kinematic control solution for multiple cooperative manipulators in a pre defined trajectory with object manipulation, while a depth vision system locates and tracks an object using Particle Filter. The general objective is to present a solution for task space control with kinematic control for multiple manipulators, with the ability to locate and track the position of the manipulated object. To this end, kinematic control techniques are used with an approach for multiple cooperative robots. Three techniques for jacobian inverse are used, being them jacobian inverse, pseudo inverse and damped least squares. The control solutions used for the manipulators are proportional control, proportional and integral control and proportional feedforward control. These techniques are implemented in the V-REP simulator and in an experimental enviroment with two UR5 manipulators. Step response and pre defined trajectory control tests are realized. For object location and tracking, a Kinect is used in the ROS ambient with PCL. A technique for localization based on models with particle filter is used. Experiments with both arms manipulating an object are done, using cooperative manipulators principles and the kinematic control techniques developed. Both systems are integrated together for dual arm kinematic control with object localization and tracking. The jacobian inverse techniques developed are applied with success and a comparison beetween the performance of the three is done, where it was found that the damped least squares method is the best method for the rest of the work. The trajectory control is successfully done with the three controllers, but the steady state error goes to zero only with the proportional feedforward control. The dual arm manipulation is done with success, but there is a delay in the communication system, resulting on a steady state error. The model based localization method developed can locate and track the object during the trajectory but, because a uncertainty in the experimental enviroment and sensor inaccuracy, the tracking quality is smaller than expected. Nevertheless, the tracking is done successfully.