Deep Reinforcement Learning Controller for Vision-Based Serial Flexible Link Manipulator

In recent years, Flexible Link Manipulators (FLMs) find a wide spectrum of applications including space exploration, defense and medical services owing to several advantages over the rigid manipulators. However, due to the flexible structure of the links in these manipulators, a number of control complexities arise. Owing to non-collocated sensors and actuators, FLM acts as a non-minimum phase system. Therefore, it is challenging to design a control scheme to achieve perfect tip tracking performance with a small tracking error. The objective of this study is to design adaptive intelligent tip-tracking control strategies for FLMs. A vision sensor is used along with a standard mechanical sensor to provide an indirect measurement of tip point deflection. The last decade witnessed a great deal of research interest in visual servoing (VS) based control of FLM. To deal with the Field-of-View (FOV) issue of conventional Image-based Visual Servoing (IBVS) control scheme an intelligent Vision-based (IVB) controller with Deep Reinforcement Learning (DRL) is developed for tip-tracking control of FLM. In this paper, the performance of the designed controller is investigated using simulation studies. It is found that the proposed controller is able to quickly correct the tip position to bring the object within FOV to complete the visual servoing task.