Dual Artificial Potential Field (d-APF) Based Control for Safe and Efficient Robotic Operation on Industrial Collaborative Scenarios.
Diego Rodríguez Guerra
- DIRECTORS: Gorka Sorrosal Yarritu e Itziar Cabanes Axpe
- UNIVERSITY: UPV/EHU
This Ph.D. proposes two main contributions for the safe and efficient collaborative manufacturing on modern industrial environments. On the one hand, a novel d-APF (dual Artificial Potential Field) controller is proposed to avoid simultaneously singularities and collisions while operating in a shared environment. This contribution leans on the second novelty proposed in this Ph.D., a decoupled kinematic model for non spherical wrist cobots. Without the last contribution or any similar singularity characterization, the d-APF controller cannot be implemented due to the lack of a set of characterized singular configurations.
With the aim of explaining the controller, the initial chapters of this document explain the theoretical fundamentals to implement the novel kinematic model, the reference controller and the proposed one. The main advantage of the decoupled kinematic model proposed relies in the spherical wrist technique to obtain a quasi spherical wrist, which is a non-spherical wrist that kinematically behaves as if it was spherical. Based on this new kinematic model, a set of closed solutions for the inverse kinematics and the singularities are obtained. Indeed, the joint dependent singularity characterization obtained is the key to implement the proposed d-APF controller, enabling the measure of the closeness to the singularity to compute a repulsive response and a limiting velocity index for each singular joint. In addition to the controller and the kinematic model, the reference or benchmarking controller is also implemented, the DLS-APF controller. This controller is not only utilized as a reference controller that integrates the singularity handling throught a DLS kinematic model, but also sets the basics for the collision avoidance components of the d-APF controller.
Then, the implementation of the model and both controllers is regarded for a UR10e robot and the SUPSI robot. In the case of the UR10e robot, the manipulator behavior for both controllers has been tested for both, a combined Gazebo-MoveIt simulated environment and real demonstrator environment. Complementary, the SUPSI robot has adopted a more supportive role to verify and extend the conclusions obtained from the first robot experimentation to other non-spherical wrist cobots with similar structure. Therefore, only simulation behavioral tests have been executed for this robot. This implementation for both robots relies on a generic software control architecture developed in this Ph.D. that creates a RTT Controller Manager and a hardware interface bridge to integrate ROS Control capabilities into Orocos components. On top of that, a specific RTT ROS Controller is implemented for each robot and each studied controller (the DLS-APF and the d-APF ones), achieving a hardware agnostic real time capable ROS controller.
Subsequently, several tests on the implemented controller have been implemented to test the real time and collaborative application suitability of the proposal. The results shows an improvement of a rough 17% in the computational time to compute a singularity free collision avoidance response with respect to the reference controller. Moreover, the real time encapsulation of the ROS Control capabilities allows sending commands to the robot at frequency rates above 500Hz, enabling the response of the robot to obstacles in very short times. It means that, depending on the application requirements, the determinism of a safe response can be assured favouring the real time capabilities of the robotic control system. Finally, some
trajectory execution tests have been executed with and without obstacles, ending with positive results where a reduction of about the 40% of the required execution time have been appreciated compared to a fully manually developed strategy. Therefore, the proposed approach after a thoughtfull analysis has been determined as suitable to improve current manufacturing efficiency on collaborative scenarios (roughly 11% of reduction in time) without disregarding the safety aspects due to the simultaneous collision and singularity avoidance that allows evading collisions in a smoother manner.