Accurate Simulation of a Collaborative Robot Arm with Cartesian Impedance Control

Abstract: Simulation of systems is used in several fields of science as a tool for safe and resource-efficient testing, as well as a tool for prediction. In this thesis, the goal is to produce an accurate simulation of a collaborative robot arm, together with a controller solution. The robot is supposed to learn and perform contact-rich tasks. Impedance control is often the suggested control strategy for such tasks, since this type of controller relates kinematics with dynamics to ensure appropriate interaction forces by enforcing the robot to behave like a mass-spring-damper system. The implemented controller for this thesis is based on a Cartesian controller called ’Forward Dynamics Compliance Controller (FDCC)’. It is using ROS and is developed to be a control solution for a wide range of different robot arms. Together with the physics engine ODE, the robotics simulator Gazebo is used as the simulation environment in this project. In this thesis, the controller framework is applied on a KUKA LBR iiwa 7 degree of freedom (DOF) lightweight robot arm. This project is the first known application of the FDCC as a controller to a 7 DOF arm and on a robot without a dedicated end-effector force-torque sensor. Instead of using an end-effector force-torque sensor, the KUKA robot uses an embedded observer for external joint torques to measure interaction forces. In this project, effort is put into processing sensor signals and tuning of control parameters to make the FDCC operate on the KUKA robot. The signal processing consists of frequency filtering and by limiting the rate of change on the sensor signal. Further on, using the same Cartesian impedance controller, the controller parameters are tuned with the result of a stable behaviour and a simulation response closely matching real system response for commanded Cartesian trajectories. Physical Human-Robot Interaction tests also show stable and responsive behaviour. As a final application example, a peg-in-hole insertion task is solved by both a simulated and a real system.