Missing training data and model weights for multimodal training pipeline

[gogojjh]

Thanks for your great work. I have several questions:

1. Missing Training Data

Training scripts expect processed data that doesn't exist:

# train_score.sh expects:
../../robot_arm_data/process_data/actor_multistep_one/

# train_critic.sh expects:
../robot_arm_data/process_data/open_lever_pull_right_critic_multistep/

# train_actor.sh expects:
../../robot_arm_data/process_data/actor_multistep_one/

Questions:

• Should these datasets be provided, or do we need to generate them?
• Is there a script to process raw simulation data (e.g., logs/_seed2/lever_handle_one/2/data_2.pt) into the expected format?
• How many simulation runs are needed for training?

2. Missing Pre-trained Weights

No pre-trained model weights are available for ScoreNetwork, CriticNetwork, or ActorNetwork.

Questions:

• Are pre-trained weights available for download?
• If yes, how to setup the code to implement the trained policy network.

Thanks!