A Deep-Learning Approach for Computing Adaptable Grasp Poses in Human-Robot Collaboration

In the field of robotics, robot grasping generation is a critical area of research aims at enabling robots to manipulate objects autonomously and effectively. The ability to grasp objects properly is fundamental for a wide range of applications, from industrial production lines to human-robot collaboration. In this context, the ability to generate adaptable grasps becomes particularly important when robots need to operate in complex and dynamic environments, often in collaboration with humans. In this work, we present a deep learning-based approach for generating adaptable grasp poses in human-robot collaboration scenarios. Our proposal relies on deep neural networks trained on different datasets of common objects to generate grasp poses that are suitable for human-robot collaboration. We will illustrate the details and the results of the proposed architecture and evaluate its performance through a series of simulated experiments.