Force control of robotic manipulators in cooperation

Cooperation of robotic manipulators has been an important field of study in robotics. Where initially, robotic design, study, and performance was targeted toward accomplishing simple tasks, more recent emphasis is placed on accomplishing complex objectives that often require active cooperation of multiple robotic units. With increased presence of robotic devices, effective cooperation opens possibility for greater automation of various tasks. With proliferation of affordable actuators, controllers and sensors, more and more robots make their ways not only onto factory floors and production lines, but also into our homes. The effective application of robotic cooperation leads right to that of human-robot cooperation. Here, robots can be used to assist human directly. A person acting as a master, as a leader, or even as a collaborator, can work side by side with a robot. In the following pages, cooperation between master and a slave is explored. Master unit is tasked with executing predefined motion while carrying a load. The slave unit assists by sharing the weight of the item while following motion path projected by the master. Both units are separate and no data exchange takes place between two robots. The compliance of the slave manipulator is based strictly on the load feedback from the wrist mounted load cell. Load feedback is also used by the master to compensate for the reaction loads arising during cooperation. The ultimate goal of the thesis was to implement above scheme on the actual hardware available at the Mechatronic Education and Research Center. Three Stäubli Rx160/Rx160L robotic units were available for testing. These are six axis robotic arms. All joints are rotary and are driven by brushless motors coupled with resolvers, and contain integrated parking brakes. Two of the robots are equipped with six-degree force/torque cells allowing for load feedback necessary for the implementation of the force control. Initial research included literature survey. Previous work concerning cooperation of the Master-Slave couple, dating as far back as early 1970s, was analyzed. Theoretical background necessary for a proper description of the system was studied. While the computed torque control, and the direct force control structures are both well-developed subjects, the actual implementation is more difficult. Modeling and simulation of the robot kinematics, dynamic and control was performed as the preliminary step for the implementation on the actual hardware. Model of the master robot was based on the inverse dynamics, computed torque control with PD position and orientation control loops. Slave unit uses direct force/torque PD control. Both use Newton-Euler algorithm for the calculation of the required joint torques. Matlab Simulink (R2014a) was used to simulate individual control structures for the master, the slave, as well as for the combined system. Coupling between master and slave manipulators was done using ideal model of a linear and of torsional springs. Linear and angular displacements between both end effectors were tracked and used to generate loads. Path planning for the master robot was based on via-point trajectory generation in task space using quintic splines. Ultimately, simulation studies allowed verification of the control algorithms and as a way to familiarize one with complete control structure. Furthermore, initial estimation of the required gains was achieved. Experimental work began once satisfactory performance of the simulated system was obtained. Matlab code was rewritten into C language to be used in the controller. Using test software, control algorithm was coded and tested before uploading it onto the robot. This was done to ensure safety. While the opening performance left a lot to be desired, subsequent developments brought increased reliability. Necessity of good characterization of the system was recognized early. As the dynamic performance of the feedback linearization strategy relies heavily on cancelation of nonlinear terms, good friction model, among others, was necessary. While positional control could achieve relatively good results, orientation control was virtually nonexistent without proper friction model. Due to very low inertia of the fifth and the sixth links friction forces dominated. Good friction model was then paramount to even actuate last two joints. Friction model, borrowed from the work of Waiboer (2007) has been adjusted for better performance. Additionally, characterization of robot's torsion spring, integral component of the second link, was done as discrepancies arose between available model and actual hardware. Gain tuning was completed using Ziegler – Nichols method for a PD controller. Additional, manual tuning was done afterwards. As a way to further improve performance, integral control action was added. It has been noted, that both control loops, position and orientation control, are often in direct conflict. As such, there is often a tradeoff in terms of achievable positional and orientation control. Increasing one leads directly to deterioration of the other. Force control of the slave manipulator was next to be implemented. While the same Newton-Euler algorithm was used, friction model was no longer crucial. Early experiments failed due to pre-preprogrammed tool frames in the load cell controller. Once corrected, performance improved dramatically. Further gain tuning achieved good response. Gripper weight at any orientations was compensated, and a separate algorithm was used to calculate weight of the held objects using force feedback form the load cell. Two control modes for the slave were realized. In addition to pure compliance control, compliance control with orientation control was added. In the first mode, end effector would actively change orientation to eliminate force/torque error between desired and feedback loads. In the second mode, orientation loop was active and would attempt to hold initial orientation of the gripper while moving in response to external loads. First cooperation experiments were performed with a human in place of a Master. Later, coupling between master and slave was accomplished by linking them with a compression spring. Experimental results are given. Following pages' detail work outlined above. It is shown that the Master-Slave cooperation allows for simple yet effective coupling of two robotic manipulators. As both arms execute separate routines and no information sharing takes place, more slave units can be coupled to accomplish a given task. Furthermore, it is shown that the master unit can easily be replaced by a human being further extending applicability of the presented work.