search

haosulab / ManiSkill

SAPIEN Manipulation Skill Framework, a GPU parallelized robotics simulator and benchmark
https://maniskill.readthedocs.io/en/latest/
Apache License 2.0
561 stars 97 forks source link

Unable to train 'PickclutterYCB' using PPO #387

Closed jayaramreddy10 closed 2 weeks ago

jayaramreddy10 commented 2 weeks ago

Hi,

I am facing an issue when training PPO policy on 'PickClutterYCB' task.

Command used: python examples/baselines/ppo/ppo.py --env_id="PickClutterYCB-v1" --exp-name="state-pickclutterYCB" --num_envs=1024 --update_epochs=8 --num_minibatches=32 --total_timesteps=5_000_000 --eval_freq=8 --num-steps=20

Issue: Traceback (most recent call last): File "/home/jai/ManiSkill/examples/baselines/ppo/ppo.py", line 270, in returns = np.concatenate(returns) ValueError: need at least one array to concatenate

Am I using the wrong command line args for training this skill?

StoneT2000 commented 2 weeks ago

This occurs if the number of evaluation steps is too small and so none of the episodes complete during evaluation.

Try setting num_eval_steps=100 (it has to be at least as high as the max episode steps value of the environment).

jayaramreddy10 commented 2 weeks ago

Got it, thanks a lot. Its working now.