Abstract
We present an implementation of a biologically inspired model for learning multimodal body representations in artificial agents in the context of learning and predicting robot ego-noise. We demonstrate the predictive capabilities of the proposed model in two experiments: a simple ego-noise classification task, where we also show the capabilities of the model to produce predictions in absence of input modalities; an ego-noise suppression experiment, where we show the effects in the ego-noise suppression performance of coherent and incoherent proprioceptive and motor information passed as inputs to the predictive process implemented by a forward model. In line with what has been proposed by several behavioural and neuroscience studies, our experiments show that ego-noise attenuation is more pronounced when the robot is the owner of the action. When this is not the case, sensory attenuation is worse, as the incongruence of the proprioceptive and motor information with the perceived ego-noise generates bigger prediction errors, which may constitute an element of surprise for the agent and allow it to distinguish between self-generated actions and those generated by other individuals. We argue that these phenomena can represent cues for a sense of agency in artificial agents.