Setting up seed properly in Custom env

[Chainesh]

Question

`# Create the training environment
train_env = gym.make("BabyAI-GoToLocal-v0",max_episode_steps= 512)#, render_mode = "human")
train_env = RGBImgPartialObsWrapper(train_env)
train_env = CustomEnv(train_env, mission)
train_env = ImgObsWrapper(train_env)
train_env = Monitor(train_env)
train_env = DummyVecEnv([lambda: train_env])
train_env = VecTransposeImage(train_env)

Create the evaluation environment

eval_env = gym.make("BabyAI-GoToLocal-v0", max_episode_steps= 512)#, render_mode = "human")
eval_env = RGBImgPartialObsWrapper(eval_env)
eval_env = CustomEnv(eval_env,mission)
eval_env = ImgObsWrapper(eval_env)
eval_env = Monitor(eval_env)
eval_env = DummyVecEnv([lambda: eval_env])
eval_env = VecTransposeImage(eval_env)

save_path = f"main_code/New_model/PPO/{mission.replace(' ', '_')}_model"
eval_callback = EvalCallback(eval_env, callback_on_new_best=stop_callback, eval_freq=8192,
best_model_save_path=save_path, verbose=1, n_eval_episodes= 30)

model = PPO("CnnPolicy", train_env, policy_kwargs=policy_kwargs, verbose=1,
learning_rate=0.0005, tensorboard_log="./logs/PPO2/",
batch_size= 2048,
n_epochs= 100, seed = 42)

model.learn(2.5e6, callback=eval_callback)

Close the environments

train_env.close()
eval_env.close()`

I'm trying to train babyai env on specific instructions, but while setting up the seed it gets stuck on first frame and doesn't do anything. The issue seems to be with Custom env because when I run the code without it, the seed works fine. Help from the community would be much appreciated :)

[pseudo-rnd-thoughts]

This appears like a sb3 issue rather than Gymnasium, could you clarify what the Gymnasium issue is?