Zeid El Idani, Elektroteknik

​Title: Sample-Efficient Search for Reactive Grasping using Fingertip Force/Torque Sensors

Supervisor: Yasemin Bekiroglu
Examiner: Yiannis Karayiannidis
Opponents: Erik Carlbaum, Martin Adolfsson

Abstract:
Success of a grasping and manipulation system depends on various factors, starting from the perception and planning stage, to the end of a task completion. Robustness during grasp execution is crucial for successful systems, which can be highly effected by many sources of uncertainty inherent in real-world settings, such as imperfect pose estimation, unknown friction or deformability properties of objects. Most grasping approaches mainly only apply predefined fixed forces during grasp execution based on simplifying assumptions on the objects, i.e., that the forces would not cause damage or failure.

However, this cannot be guaranteed in the case of unknown objects. This thesis addresses the issue of grasping novel objects and is focused on increasing robustness in grasp execution using real sensory data, i.e., force/torque readings. Here, an outline of a sample-efficient approach to robust grasp execution to increase grasp success rates is given. Bayesian optimization is used for sample-efficient search, which allows for finding good grasp control parameters in a small number of trials using a real robot. The results from real robot experiments show that the proposed approach can be successfully applied to grasp execution, where unknown objects can be grasped without any damage or slippage while passing stability tests.

Welcome!
Zeid, Yasemin and Yiannis
​Join the seminar by Zoom
Password: 1234
Kategori Studentarbete
Plats: Landahlsrummet, conference room, Hörsalsvägen 9, EDIT trappa F, G och H
Tid: 2022-06-14 13:00
Sluttid: 2022-06-14 14:00

Sidansvarig Publicerad: to 02 jun 2022.