notmahi / bet

Code and website for Behavior Transformers: Cloning k modes with one stone.
https://mahis.life/bet
MIT License
109 stars 16 forks source link

Data Description of Kitchen Franka #10

Open liubaoryol opened 2 weeks ago

liubaoryol commented 2 weeks ago

Hi! I'm trying to make sense of the data that you have used in detail. I have looked into Franka-Kitchen documentation however there are some differences that I would like to understand. For instance, it has an observation of 59 instead of 60 , and also the code of this repo has the following block in envs/kitchen/__init__.py:

OBS_ELEMENT_INDICES = {
    "bottom burner": np.array([11, 12]),
    "top burner": np.array([15, 16]),
    "light switch": np.array([17, 18]),
    "slide cabinet": np.array([19]),
    "hinge cabinet": np.array([20, 21]),
    "microwave": np.array([22]),
    "kettle": np.array([23, 24, 25, 26, 27, 28, 29]),
}

Where can I find the meaning of each entry of observation?

Thank you so much!

jeffacce commented 2 weeks ago

Hi there!

If I recall correctly, the first 30 elements contains the environment state, and the last 30 elements contains the goal state I believe. The first 9 elements should be the robot states, and elements [9:30] should be object states I think.

Also see this in d4rl: https://github.com/Farama-Foundation/D4RL/blob/cf0efbd75a9e05f34dfee2f566e163abfa512485/d4rl/kitchen/adept_envs/franka/kitchen_multitask_v0.py#L130