I started my Ph.D. by developing an Active Simultaneous Localization and Mapping (SLAM) system to be used in an indoor environment. The system is made of a three-layered framework that can work for both omnidirectional robots as well as other ground platforms. The framework, called iRotate, exploits active vision and concepts for both long- and short-term exploration techniques. This is done by continuosly optimizing the camera orientation during the experiment and along the planned trajectory. When the robot is not omnidirectional, control the camera through an independent camera rotation mechanism. This required a novel state estimation approach to avoid accumulating camera rotation and robot movements errors. With this, I can autonomously explore an environment with up to 39% shorter paths with a substantial decrease in energy consumption --- a critical aspect in mobile autonomous robotics. I tested the system first in simulation and then validated the results with a real robot.

The main limitation of the iRotate is that it can only work in static environments. However, simulating realistic dynamic environments with robotics-enabled tools was not an easy task.

For that reason, I developed GRADE, a framework built over IsaacSim that can solve this problem. With that, I demonstrated the limitation of the current state-of-the-art methods in dynamic SLAM, which either fail to track a trajectory fully or do that with big estimation errors. Moreover, I show how the data generated can be used effectively for vision related tasks such as the detection of humans or zebras from aerial views.

Within the Robot Perception Group, I worked in multi-agent systems for aerial motion capture using perception (AirCap) and Deep Reinforcement Learning (AircapRL). I also participated in the AirPose project contributing to the data generation and the synchronization mechanism of the real drones.

Previously, I've obtained my B.Sc. in Computer Science Eng. (2013) and M.Sc. in ICT for Internet and Multimedia Eng. (2019) at the University of Padua in Italy. Shortly after, I started working as a research intern at MPI-IS in Tuebingen, Germany.

active SLAM SLAM synthetic data generation 3D reconstruction active vision robotics