Enabling Dynamic Reconfiguration of Numerical Methods for the Robotic Motion Control Task

• Published in: 2016 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW) pp. 283 - 288
• Main Authors: Schwiegelshohn, Fynn, Kastner, Florian, Hubner, Michael
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.05.2016
• Subjects: Hardware; Jacobian matrices; Kinematics; Manipulators; Robot kinematics; Service robots
• DOI: 10.1109/IPDPSW.2016.131

Current industrial robots are increasing in significance as they are able to autonomously perform more and more complex tasks. These tasks need to be executed fast and accurate. For environment interaction, the inverse kinematics problem needs to be solved. This problem becomes increasingly difficult when the robot has many degrees of freedom. Therefore, industrial robots are specifically designed to enable an analytical solution of the inverse kinematics problem. Through these design restrictions, redundant robot designs cannot be realized. Redundant robots are more flexible in their movement and are better able to perform tasks with obstacles in their workspace. Numerical methods are able to solve the inverse kinematics problem without restricting the robots design. But, numerical methods have their strengths and weaknesses in different movement situations. In order to cover all possible situations satisfactorily, a combination of several numerical methods is required. In a software implementation, this leads to long processing times that are not feasible in a real-time environment. By implementing numerical methods on reconfigurable hardware, the processing time can be reduced and several numerical methods can be dynamically loaded onto the hardware if the current situation requires a different numerical method for maximum performance. Therefore, this paper analyzes the inverse Jacobian, the pseudo-inverse Jacobian, the gradient descent, and the Newton method and implements these numerical methods on reconfigurable hardware. Furthermore, the hardware architecture and all numerical methods are designed to support dynamic reconfiguration. Our system is evaluated with the robotic arm platform AREXX RA1-PRO. The numerical methods show good performance when implemented on reconfigurable hardware. Dynamic reconfiguration is also possible although its true benefit becomes apparent when obstacle detection and avoidance is also implemented on a higher level.