What Do I Look Like? A Conditional GAN Based Robot Facial Self-Awareness Approach

In uncertain social scenarios, the self-awareness of facial expressions helps a person to understand, predict, and control his/her states better. Self-awareness gives animals the ability to distinguish self from others and to self-recognize themselves. For cognitive robots, the ability to be aware of their actions and the effects of actions on self and the environment is crucial for reliable and trustworthy intelligent robots. In particular, we are interested in robot facial expression awareness by using action joint data to achieve self-face perception and recognition, passing a deep learning model. Our methodology proposes the first attempt toward robot facial expression self-awareness. We discuss the crucial role of self-awareness in social robots and propose a CGAN (Conditional Generative Adversarial Network) model to generate robot facial expression images from motors’ angle parameters. By using the CGAN method, the robot learns its facial self-awareness from a series of facial images. In addition, we introduce our robots facial self-awareness dataset. Our methodology can make the robot find the difference between self and others from its current generated image. The results show good performance and demonstrate the ability to achieve real-time robot facial self-awareness.