This paper has received 3 favorable reviews, indicating that the paper addresses an important problem (learning affordances from explorations and interactions) in a novel way through learning of segmentation model to the classical RL problem. The reviewers laud novelty, comprehensive evaluations, empirical results. The rebuttal could address several points raises by reviewers, e.g. noisy odometry. The strengths of the paper clearly outweigh the perceived weaknesses, and the AC concurs.