PandaPush Sim-to-Sim Transfer Project

This folder contains only the files required to train and evaluate the final PandaPush-v3 policies.

Structure

Reinforcement_Learning_Project/
|-- train_sb3.py
|-- eval_sb3.py
|-- rand_wrapper.py
|-- requirements.txt
|-- CONFIGURATIONS.md
|-- models/
|   |-- ppo_push_none_source_2000k.zip
|   |-- sac_push_none_source_500k.zip
|   |-- sac_push_none_target_500k.zip
|   |-- sac_push_udr_source_500k.zip
|   `-- sac_push_adr_source_500k.zip
`-- results/
    |-- training/
    `-- evaluations/

Roles of the Python files

• train_sb3.py: trains PPO or SAC using a fixed mass, UDR, or ADR.
• eval_sb3.py: evaluates a saved policy on the fixed source or target mass.
• rand_wrapper.py: applies the object mass and implements UDR/ADR.

Training logs and configurations are automatically written to results/training/. Evaluation CSV files are automatically written to results/evaluations/.

See CONFIGURATIONS.md for installation, training, evaluation, and rendering commands.