Low-cost vision-based embedded control of a 2DOF robotic manipulator

Simulations are frequently used to validate underactuated control of the planar twolink manipulator, a typical benchmark system. These new control strategies need to be evaluated on a mechanical setup in order to further research their validity. In this work the benchmark system is specifically designed and built to fit on a table and to be as low-cost as possible. One of the large discrepancies between the simulator and the real setup is the presence of friction. With the help of a carefully designed bearing topology and by using image vision to measure the states of the system, as opposed to mechanical sensors, this friction can be kept to a minimum. The image vision requires a computer that is capable to process the images and send the control effort to the actuator in a timely manner. For this purpose a Raspberry Pi 4 is used to interface with the camera. The architecture consists of a traditional Raspbian operating system, which is expanded with Robot Operating System (ROS) and Xenomai. With the latter, tasks can be done in hard real-time, which is needed to achieve a fixed sampling time. In this case, a sampling time of 9ms is reached. An open-loop and closed-loop experiment with PD-controller are performed to validate the software architecture.