Closed devinluo27 closed 1 month ago
Hello, I am looking into this. Could be some non-robust behavior I accidentally created when cleaning up the code
Sorry about the issue, this is confirmed to be a bug when cleaning up the code. I will push the fix among a much improved version with my new transformer implementation of df_planning next week, with much better results and much faster speed!
Thank you for the update! I will check out the updated version once it's published.
It seems I need more time to release a nice and clean repo for the transformer version of planning. However, I fixed this problem for now in #2, just follow the instructions in it and try my ckpt!
I just released the transformer version on main branch. Check it out and here are some visualizations of it!
I trained diffusion forcing on Maze2d following the command in README. When I do the planning using the trained model, the output actions are 'NaN' occasionally, which compromises the mujoco simulation. Do you have any suggestions to resolve this issue?