Unable to Replicate Results of Fig. 3a and Fig. 3b (for MetaWorld benchmark) #27
Description
Hi, thanks for code release. I am currently unable to reproduce the results reported in Figure 3a and Figure 3b of the paper (Metaworld benchmark). Despite following the provided training and evaluation scripts, my reproduced success rates are lower than those shown in the figures.
Observed Results:
Multi-task success rate: ~84.3% (91.7 reported in paper)
Few-shot transfer success rate: ~38.96% (71.9 reported in the paper)
Please let me know if there are any differences between the released code and the setup used for the reported figures?