Autonomous Robotic Exploration based on Visual Data

This work focuses on autonomous 3D coverage of an environment using a camera placed on a mobile robot base. In this computer vision scenario, coverage is intimately related with visibility. This work does not address a particular 3D reconstruction technique but it is more focussed on the definition of custom policies for the coverage of a 3D area having a vision-based acquisition sensor which dynamically explores the environment. The mobile platform is modelled in ROS and we ask it to obtain the "best" possible coverage by exploring the area with certain criteria (i.e. trying to maximize the informations acquired while minimizing the associated costs). The main issues faced in this work are how to define properly the concept of current coverage for a given portion of the map and how to define a reward function that guides the exploration taking into account both the utility, in terms of increased degree of coverage acquired, and the cost for the robot to move on the map. We have tested our model on a particular 3D reconstructed environment which is Piazza Bra in Verona (Italy). This environment represents our starting point as it allows the robot to simulate a real exploration.