Robot emotions generated and modulated by visual features of the environment

Emotions are generated and modulated by many factors in the ever-changing surrounding environment. A new and challenging task is to emulate emotional responses on a robot that are caused by visual stimuli, such that the robot´s responses mirror that of the human user. This paper presents the initial stage of an affective system that has been trained on-line using reinforcement learning to generate and modulate emotions. The inputs of the system comprise a subset of emotionally relevant visual features extracted from the environment: colours, fractal dimension, and facial pareidolia. These inputs are mapped onto an output that expresses the associated emotion in terms of language. Pilot experiments demonstrate how a humanoid robot tries to learn through interaction with a human companion to express emotions associated with different environmental scenes in a (near) human-like manner.