Fast predictive visual servoing: A reference governor-based approach

In the robotic domain, the well-known advantages of using predictive controllers have to be carefully balanced with the fast execution needs of the low-level motion control loops. In this work, we propose a control architecture with a reference governor to enhance visual servoing with a predictive behavior, without giving up on the fast execution of the original reactive controller. The reference governor, implemented as a model predictive control, acts as an outer loop running at a relatively low frequency over the original visual servoing loop. This strategy takes advantage of the model prediction to improve the performance of the reactive controller and makes the whole structure more flexible. The proposed approach has been validated through simulations with a 7-axis robotic manipulator, a humanoid robot, and a hybrid wheeled-legged quadruped robot. Experiments on a real 7-axis manipulator show the validity of the approach in real-world scenarios.