Human hand motion embodiment mapping from RGB images to robotic hands

Policy development of fine motor control commands of robotic grippers, especially with a high number of degrees of freedom, involves complex training sessions that are practically impossible outside of laboratories without expert robotics knowledge.  A fully combined software solution that allows robots to recognize human hand motions from RGB images and translate them into robot gripper poses would be an important first step that can simplify teaching robots. However, since the visual estimation of human hand motions from RGB images is a complex task with a relatively high error potential, the aim of this master thesis is to compare the accuracy of the state-of-the-art RGB perception models with that of conventional motion marker systems.