Unable to achieve good results on Humanoid Reach/Location/Strike tasks

[floodsung]

Hi Jason, I am trying to train the tasks but only Humanoid Heading works, others failed with pretty low rewards. I train the task following your script like this:

python ase/run.py --task HumanoidReach --cfg_env ase/data/cfg/humanoid_sword_shield_reach.yaml --cfg_train ase/data/cfg/train/rlg/hrl_humanoid.yaml --motion_file ase/data/motions/reallusion_sword_shield/RL_Avatar_Idle_Ready_Motion.npy --llc_checkpoint ase/data/models/ase_llc_reallusion_sword_shield.pth --headless Is there anything wrong in the config? And can you share the trained model? Thx

[xbpeng]

Yea, seems like the tasks are not working as well as I expected. There might be an issue with the retrained ASE model. Looking into it now. How many samples did you train each policy with?

[xbpeng]

The task training issues should be fixed now. I've also uploaded new pre-trained models for the low-level controllers and high-level controllers.

[floodsung]

Thank you. I have tested and it works!

[makeme-zgz]

hi, @xbpeng may I ask what was the issue with the pretrained ASE model? and what's being fixed?

[xbpeng]

It wasn't so much an issue with the pre-trained model. I just had to tweak the reward functions for the downstream tasks. Had to shape the reward a bit to give the character more incentive to move towards the target.