Pregled bibliografske jedinice broj: 1192863
Artificial Intelligence-based Methods for Interaction Force Estimation and Mediated Navigation of a Robot Manipulator on a Mobile Base
Artificial Intelligence-based Methods for Interaction Force Estimation and Mediated Navigation of a Robot Manipulator on a Mobile Base, 2022., doktorska disertacija, Fakultet elektrotehnike, strojarstva i brodogradnje, Split
CROSBI ID: 1192863 Za ispravke kontaktirajte CROSBI podršku putem web obrasca
Naslov
Artificial Intelligence-based Methods for
Interaction Force Estimation and Mediated
Navigation of a Robot Manipulator on a Mobile Base
Autori
Kružić, Stanko
Vrsta, podvrsta i kategorija rada
Ocjenski radovi, doktorska disertacija
Fakultet
Fakultet elektrotehnike, strojarstva i brodogradnje
Mjesto
Split
Datum
18.03
Godina
2022
Stranica
133
Mentor
Musić, Josip ; Kamnik, Roman
Ključne riječi
mobile manipulator, mobile robot, robotic manipulator, neural networks, deep learning, fuzzy mediation, navigation, obstacle avoidance, force estimation
Sažetak
Mobile robotic manipulators have recently emerged and are today found in industrial and domestic environments. They can perform various tasks due to their ability to move in space. Since they consist of a mobile base on which a robotic manipulator is mounted, the knowledge of both mobile robots and robotic manipulators is required to control it effectively. Thus, efficient control schemes for both mobile base and manipulator need to be developed. This dissertation tackles both fields and proposes solutions to common problems in them. First, an efficient and modular control scheme that achieves the complex behaviour of the mobile base is developed. It is achieved using fuzzy mediation to fuse two simple behaviours (navigation to a given goal and obstacle avoidance). Also, an approach to obstacle avoidance based on neural networks is developed and successfully incorporated into the fuzzy mediation scheme. In the experiments, the approaches demonstrated good performance in navigation on two real-world robots of different sizes and shapes. Furthermore, approaches (based on neural networks) are proposed to estimate the end-effector forces acting on the manipulator and to estimate joint-side torques by using the sensor mounted under the robot base. The networks were trained for two manipulators of different sizes and payloads. They generalised well to unseen trajectories, and force and torque estimates are reasonably accurate.
Izvorni jezik
Engleski
Znanstvena područja
Elektrotehnika, Računarstvo
POVEZANOST RADA
Ustanove:
Fakultet elektrotehnike, strojarstva i brodogradnje, Split