Figure 1: Evaluation with added comparison to PEARL, showing meta-training curves on full state pushing (left), ant locomotion

However, GMPS significantly outperforms PEARL on sparse reward tasks. GMPS is better able to learn out-of-distribution tasks. Ablation for number of consecutive outer updates, as requested by reviewer 3. Using 500 imitation steps (blue) We thank the reviewers for their positive and constructive feedback. The primary concern from Reviewer 1 was the comparison to PEARL (Rakelly et al.). Reviewer 1. See PEARL comparisons above.