The strength of the paper lies in proposing a novel method which is interesting, simple, novel, yet effective. On the other hand, the reviewers were concerned that the proposed approach is somewhat heuristic, and it is unclear why it works and what its limitations are. Specifically, the reviewers were unclear why the method helps in the corridor and u-shaped maze experiments; a much more thorough analysis of these cases would be greatly beneficial to the reader. We recommend to add additional baselines (such as a grid-oracle baseline), experiments (such as adding noise to the starting condition), and analysis of results, as suggested by reviewers. The authors should also clarify some points in the "self-balancing reward" section that were unclear to the reviewers.