HybGrasp: A Hybrid Learning-To-Adapt Architecture for Efficient Robot Grasping

away more precise details of the gripper, such as finger size, finger arrangement, and grasping trajectory. Even if the approximate angle-width prediction of the model is correct, this limitation can lead to a failure case in which a collision occurs as the gripper tries to pick up an object because the model does not take into consideration the specifications of the gripper itself [11]. In this letter, we propose a novel robot-grasping framework to address the problems above. Within this framework, we introduce HybGrasp, a hybrid architecture that combines a IEEE Robotics and Automation Letters (RA-L) paper, presented at ICRA 2024, Yokohama, Japan. Cite as RA-L paper. IEEE Robotics and Automation Letters (RA-L) paper, presented at ICRA 2024, Yokohama, Japan. Cite as RA-L paper.