We add few lines to add an extra level-2 minibatch for Diffusion Policy (as per Mini Diffuser (ICRA 2026)) trained for the PushT environment from gym-pusht. This enables tens of equivalent batch size per gradient step, and end up saving at least 60% of the training time to obtain similar training results.

See the LeRobot library for instructions on how to load and evaluate this model.

Trained with a forked LeRobot@ 7bd533a.

The model was trained using LeRobot's training script and with the pusht dataset, using this command:

The training curves, and the comparasions with original DP may be found at https://api.wandb.ai/links/hu2240877635/defcr4wu