Position Based Visual Control of the Hovering Quadcopter

Autonomous navigation of quadcopters in unstructured indoor environments is a major problem due to the difficulty of reliable position sensing. While outdoor applications can use GPS for reliable localization, working indoors will require the use of either laser range finders or some other sensors. If the indoor scene is unknown to a robot, the task of mapping new areas also becomes a necessity. The two processes are combined and run together in a framework of Simultaneous Localization and Mapping (SLAM). Our work is focused on using onboard cameras for the task of SLAM in an indoor scenario. Vision based techniques that do not use time of flight methods like laser range finders, have the potential to provide a low cost alternative framework for navigation. In this work, localization using a monocular SLAM framework on an unknown and unstructured scene, a cascaded position controller along with a Luenberger observer which can combine the data of Inertial sensors and vision based position to generate a complete velocity feedback for the system have been used. Sensor data fusion using EKF (Extended Kalman Filter) have been performed for scale estimation. The localization algorithm has been implemented on a quadcopter. Finally hovering experiment has been performed in an indoor lab based environment.