Dataset Size

[alfredgu001324]

Thank you for your great work! It is very interesting, and I have successfully run through the pipeline.

I have a few questions:
(1) How many scenarios (or maybe episodes) are generated for each simulator to reach the target performance (whether off policy pretrain or on policy)?
(2) I have trained a scrap model, which is only trained on 50 episodes of mujoco data. And I evaluated its performance using XLA_PYTHON_CLIENT_MEM_FRACTION=0.3 ros2 launch car_ros2 car_sim.launch.py. The car goes backward lol, but I think it's normal. I am wondering is this the only way to test the model performance, or is there any other scenarios/scripts to test in simulation?
(3) When training, do you put all the pkl files into one single large folder and then train, or train separately? For example, first train on mujoco, save the checkpoint, and then load it and train on issac gym etc.

Thank you again for your great work and looking forward to your reply!