Found it very interesting. I've made several attempts to reproduce the results in your paper, but I'm currently unable to match the reported performance. Could you please provide the checkpoint file and the low_dim data containing the waypoint info used to generate the results reported in the paper?