Error when running run_oracle.sh: SubprocVecEnv has no attribute 'env'

[songyuc]

Environment

• OS: Linux
• Python: 3.7.16
• GPU: NVIDIA GeForce RTX 2080 Ti
• CUDA: 12.4
• Dependencies installed according to README.md (lines 22-41)

Description

I'm a student new to robotics and trying to run the Text2Reward project. When executing bash run_oracle.sh, I encountered the following error:

Traceback (most recent call last):
  File "ppo.py", line 161, in <module>
    success = np.array(ep_lens) < eval_env.env.env._max_episode_steps
AttributeError: 'SubprocVecEnv' object has no attribute 'env'

The error occurs when trying to access the _max_episode_steps attribute through nested env attributes of the vectorized environment.

Steps to Reproduce

1. Set up environment following README.md instructions
2. Run bash run_oracle.sh

Additional Context

• I'm using the environment setup exactly as specified in:

## Dependencies
To establish the environment, run this code in the shell:
```shell
# set up conda
conda create -n text2reward python=3.7
conda activate text2reward
# set up ManiSkill2 environment
cd ManiSkill2
pip install -e .
pip install stable-baselines3==1.8.0 wandb tensorboard
cd ..
cd run_maniskill
bash download_data.sh
# set up MetaWorld environment
cd ..
cd Metaworld
pip install -e .
# set up code generation
pip install langchain chromadb==0.4.0
```

• The vectorized environment is created as shown in:

        "# create one eval environment. Partial must be used here\n",
        "env_fn = partial(\n",
        "    make_env,\n",
        "    env_id,\n",
        "    record_dir=\"logs/videos\",\n",
        ")\n",
        "eval_env = SubprocVecEnv([env_fn for i in range(1)])\n",
        "eval_env = VecMonitor(eval_env) # attach this so SB3 can log reward metrics\n",
        "eval_env.seed(0)\n",

Questions

1. What's causing this error?
2. Could you help with solving this error?

I appreciate your help!

[sihengz02]

Apologies for the delayed response. I don’t check the issues in this repository very frequently. May I ask if this occurs only when you run bash run_oracle.sh, or does it also happen when you run bash run_zero_shot.sh or bash run_few_shot.sh?

[songyuc]

I just run bash run_oracle.sh, and have not tried bash run_zero_shot.sh or bash run_few_shot.sh yet.

[sihengz02]

Could you plz have a try and see whether you encounter the same issue?

[songyuc]

I tried running run_zero_shot.sh as suggested but encountered an error. The error appears to be related to Wandb permissions:

wandb: Using wandb-core as the SDK backend.
wandb: Currently logged in as: songyuc. Use `wandb login --relogin` to force relogin
wandb: ERROR failed to upsert bucket: returned error 403 Forbidden: {"errors":[{"message":"permission denied","path":["upsertBucket"],"extensions":{"code":"PERMISSION_ERROR"}}],"data":{"upsertBucket":null}}

Traceback (most recent call last):
  File "ppo.py", line 116, in <module>
    settings=None if args.reward_path is None else wandb.Settings(code_dir=args.reward_path[:-11]))
[...]
wandb.errors.errors.CommError: failed to upsert bucket: returned error 403 Forbidden: {"errors":[{"message":"permission denied","path":["upsertBucket"],"extensions":{"code":"PERMISSION_ERROR"}}],"data":{"upsertBucket":null}}

Even though I'm logged into wandb, it seems there's a permission issue when trying to create or update the bucket. Could you please advise:

1. Is wandb logging required for running these scripts?
2. If required, what permissions do I need to set up in wandb?
3. Is there a way to run the scripts without wandb logging?

Thank you for your help!

[sihengz02]

1. Not required.
2. You can refer to the documents of wandb.
3. Just comment the codes that involve wandb. It's quite straightforward.

[songyuc]

After a trial, I found that running run_zero_shot.sh would have a similar error.