search

MarkFzp / act-plus-plus

Imitation learning algorithms with Co-training for Mobile ALOHA: ACT, Diffusion Policy, VINN
https://mobile-aloha.github.io/
MIT License
2.98k stars 551 forks source link

eval success rate 0.5 #23

Closed lwcode closed 8 months ago

lwcode commented 8 months ago

image

task is sim_transfer_cube_scripted, 50 episodes, trainning parameters: python3 imitate_episodes.py --task_name sim_transfer_cube_scripted \ --ckpt_dir ./models --policy_class ACT --kl_weight 10 --chunk_size 100 \ --hidden_dim 512 --batch_size 8 --dim_feedforward 3200 --num_steps 2000 \ --lr 1e-5 --seed 0 --eval Is the eval result normal?

lwcode commented 8 months ago

My fault, need modify '--num_steps' to greater number.

lunalulu commented 8 months ago

My fault, need modify '--num_steps' to greater number.

Why does the success rate decrease when I increase the number of training sessions?

lwcode commented 8 months ago

My fault, need modify '--num_steps' to greater number.

Why does the success rate decrease when I increase the number of training sessions?

success rate seems to converge to fluctuate around a certain value for large num_step, for example 20000