NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:6819
Title:Control What You Can: Intrinsically Motivated Task-Planning Agent

The reviewers were in disagreement about this paper. All reviewers agreed that the method was overly complex, making it less likely that the work would have lasting impact. However, they generally agreed that the paper contained useful ideas and contributions. We encourage the authors to address the reviewers comments. In addition to the reviewers comments, it is worthwhile to note that the idea of a partition goal space has been used in the past (e.g. in the hindsight experience replay paper, contrary to R2's comment). We encourage the authors to make this connection to prior work. Further, we strongly encourage the authors to include a comparison to HIRO with the same goal space partitioning, to provide an apples-to-apples comparison regarding the problem assumptions.