deep-rl-class icon indicating copy to clipboard operation
deep-rl-class copied to clipboard

Changes to obs space descriptions in unit6.ipynb

Open Ivan-267 opened this issue 1 year ago • 0 comments

This PR proposes some changes to the description of the observation space.

However, please verify the validity first since I'm not very familiar with this env.

My current understanding after taking a look at (https://panda-gym.readthedocs.io/en/latest/custom/custom_task.html) and (https://github.com/qgallouedec/panda-gym/blob/master/panda_gym/envs/tasks/reach.py#L47-L48) is that the desired_goal is likely the target position for the end-effector (also the position of the target object used to indicate the goal) which is randomly sampled on restart, while the achieved_goal is the current position of the end-effector, with reward being -(distance between the two) in the dense version.

The success condition is then whether the difference between achieved and desired goals is less than the threshold.

Ivan-267 avatar Dec 19 '23 13:12 Ivan-267