Synthesis and Execution of Communicative Robotic Movements with Generative Adversarial Networks
Object manipulation is a natural activity we perform every day. How humans handle objects can communicate not only the willfulness of the acting, or key aspects of the context where we operate, but also the properties of the objects involved, without any need for explicit verbal description. Since human intelligence comprises the ability to read the context, allowing robots to perform actions that intuitively convey this kind of information would greatly facilitate collaboration. In this work, we focus on how to transfer on two different robotic platforms the same kinematics modulation that humans adopt when manipulating delicate objects, aiming to endow robots with the capability to show carefulness in their movements. We choose to modulate the velocity profile adopted by the robots' end-effector, inspired by what humans do when transporting objects with different characteristics. We exploit a novel Generative Adversarial Network architecture, trained with human kinematics examples, to generalize over them and generate new and meaningful velocity profiles, either associated with careful or not careful attitudes. This approach would allow next generation robots to select the most appropriate style of movement, depending on the perceived context, and autonomously generate their motor action execution.