This is a borderline paper showing how to learn a latent representation using a VAE for control of soft robots. There were concerns by the reviewers about how the results could be generalized and scaled to other simulations and systems. However, there was consensus that the work was novel and should be presented at NeurIPS.