Problems with high-level playing

[XiaoWZENG]

Hi , thanks for your impressive work. I encountered some problems when trying to playing the high-level policy. Have you ever encountered this before? Thanks a lot！

raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format( RuntimeError: Error(s) in loading state_dict for ActorCritic: size mismatch for actor.history_encoder.encoder.0.weight: copying a param with shape torch.Size([30, 71]) from checkpoint, the shape in current model is torch.Size([30, 66]). size mismatch for actor.actor_backbone.0.weight: copying a param with shape torch.Size([128, 91]) from checkpoint, the shape in current model is torch.Size([128, 86]). size mismatch for critic.critic_backbone.0.weight: copying a param with shape torch.Size([128, 89]) from checkpoint, the shape in current model is torch.Size([128, 84]).

[Ericonaldo]

what is your low-level training command and high-level training command?

[XiaoWZENG]

The low-level training command is: python train.py --headless --exptid zzc_0530 --proj_name b1z1-low0530 --task b1z1 --sim_device cuda:0 --rl_device cuda:0 --observe_gait_commands. and the low-level training policy is work for playing.

The high-level training command is: python train_multistate.py --rl_device "cuda:0" --sim_device "cuda:0" --timesteps 60000 --headless --task B1Z1PickMulti --experiment_dir b1-pick-multi-teacher --wandb --wandb_project "b1-pick-multi-teacher" --wandb_name "hight level 0530" --roboinfo --observe_gait_commands --small_value_set_zero --rand_control --stop_pick

[Ericonaldo]

Are you sure you only have this one low-level policy and the high-level assigns the right path？

[Ericonaldo]

missing dimension 5 is obviously a mismatch from observe_gait_commands, check if you are actually using the same arg for both sides.

[XiaoWZENG]

Sorry, I had set the wrong file path of the low level policy and added the observe_gait_commands argument. Now it's working. Thank you very much for your help. Have a nice day.

[whn981841576]

对不起，我设置了低级别策略的错误文件路径并添加了 observe_gait_commands 参数。现在它正在工作。非常感谢您的帮助。 有好的一天。

how to set observe_gait_commands, i ref walk these way,the observe_gait_commands is true,the same as visual whole body, how to set?