Examenspresentation av Nils Ingelhag

Supervisor: Michael Welle, KTH
Examiner: Yasemin Bekiroglu

Abstract:

Despite recent advancements robotic grasping still lags behind the dexterity and adaptability of the human hand. This project aimed to narrow this gap by developing a method to mimic human hand grasps using robotic grippers. The approach involved representing both the human hand and robotic grippers as kinematic trees, aligning these trees, and then optimizing the gripper’s pose to match that of a human hand pose. Evaluation using three robotic grippers—Shadow hand, Allegro hand, and Franka hand—on a dataset of natural human grasps revealed varying degrees of success. Grippers with higher dexterity and similarity to the human hand showed more natural alignment. However, challenges arose with the Franka hand and certain grasp taxonomies. The method’s limitations include not fully considering the kinematic model during alignment and issues with tree node weighting based on grasp type. Potential improvements are suggested to address these challenges.

Welcome!

Nils and Yasemin