Create a goal conditioned wrapper for the environments.

[virajparimi]

Hi,

I have been using the Safe Point Goal environment. However, I need to write a goal-conditioned wrapper around it for my work. I have noticed that there is a sample_goal_position function in the random_generator of the task of the environment but when i use it to sample a valid goal position it complains the following,

in sample_goal_position placements, keepout = self.placements['goal'] TypeError: 'NoneType' object is not subscriptable

I have noticed that the placements_info.placements is indeed empty so I need to initialize it. However I am not sure which API to use for that as I found the build_placements_dict function that is abstract.

Can you point me to some documentation or help me out with pointers on how can I create a goal-conditioned wrapper for the environments supported by safety-gymnasium?

[muchvo]

The specific implementation of the _build_placements_dict function is found in this file. Additionally, the sampling of the goal position information occurs within this function. Could you please provide more details about what the goal-conditioned wrapper needs to accomplish? This will help me offer more direct assistance.

[virajparimi]

Hi,

I was able to get my custom goal-conditioned wrapper implemented. The issue was that those attributes are only populated after the reset and I was not doing that before. However, I have a followup question. Right now I use the make function to generate the environment with the string ID. However, This generates an environment with bunch of prebuilt wrappers around the environment. For example it already includes the TimeLimit wrapper. I want to include my wrapper before the TimeLimit wrapper is added around the environment. Is there a way to this?

[muchvo]

1. Do you mean that you would like to customize the placement of your goal through a wrapper?

2. Yes, that's possible. Specifically, you'll need to reference the make function here. You just need to modify this function to add your wrapper. Alternatively, you could remove the automatic wrapping of the TimeLimit wrapper and define the order yourself:

   # before
   __register_helper(
       env_id='SafetyHalfCheetahVelocity-v1',
       entry_point='safety_gymnasium.tasks.safe_velocity.safety_half_cheetah_velocity_v1:SafetyHalfCheetahVelocityEnv',
       max_episode_steps=1000,
       reward_threshold=4800.0,
   )
   
   # after
   __register_helper(
       env_id='SafetyHalfCheetahVelocity-v1',
       entry_point='safety_gymnasium.tasks.safe_velocity.safety_half_cheetah_velocity_v1:SafetyHalfCheetahVelocityEnv',
       reward_threshold=4800.0,
   )
   env = safety_gymnasium.make('SafetyHalfCheetahVelocity-v1')
   env = YourWrapper(env)
   env = SafeTimeLimit(env, 1000)

Additionally, you can utilize the additional_wrappers parameter to define the order of wrappers, both before and after they are applied. For more specific guidance, please refer to this documentation. Note that this approach might require more extensive configuration.