How to use droid data and local data for co training?

[lijinming2018]

I set up:

DATA_PATH = "/home/eai/Dev-Code/droid_policy_learning"    # UPDATE WITH PATH TO RLDS DATASETS
EXP_LOG_PATH = "/home/eai/Documents" # UPDATE WITH PATH TO DESIRED LOGGING DIRECTORY
EXP_NAMES = OrderedDict(
    [
        # Note: you can add co-training dataset here appending
        # a new dataset to "datasets" and adjusting "sample_weights"
        # accordingly
        ("droid_100", {"datasets": ["droid_100"],
                   "sample_weights": [1]}),
        ("droid_local", {"datasets": ["droid_local"],
                   "sample_weights": [1]})
    ])

when I run python robomimic/scripts/config_gen/droid_runs_language_conditioned_rlds.py I get

python robomimic/scripts/config_gen/droid_runs_language_conditioned_rlds.py --wandb_proj_name ljm-2022
Num exps: 2
generated_json_paths=================== ['/home/eai/Documents/tmp/autogen_configs/ril/diffusion_policy/droid/im/04-04-None/04-04-24-14-52-45/json/bz_128_noise_samples_8_sample_weights_1_dataset_names_droid_100_cams_2cams_goal_mode_None_truncated_geom_factor_0.3_ldkeys_proprio-lang_visenc_VisualCore_fuser_None.json', '/home/eai/Documents/tmp/autogen_configs/ril/diffusion_policy/droid/im/04-04-None/04-04-24-14-52-45/json/bz_128_noise_samples_8_sample_weights_1_dataset_names_droid_local_cams_2cams_goal_mode_None_truncated_geom_factor_0.3_ldkeys_proprio-lang_visenc_VisualCore_fuser_None.json']

python /home/eai/Dev-Code/droid_policy_learning/robomimic/scripts/train.py --config /home/eai/Documents/tmp/autogen_configs/ril/diffusion_policy/droid/im/04-04-None/04-04-24-14-52-45/json/bz_128_noise_samples_8_sample_weights_1_dataset_names_droid_100_cams_2cams_goal_mode_None_truncated_geom_factor_0.3_ldkeys_proprio-lang_visenc_VisualCore_fuser_None.json

python /home/eai/Dev-Code/droid_policy_learning/robomimic/scripts/train.py --config /home/eai/Documents/tmp/autogen_configs/ril/diffusion_policy/droid/im/04-04-None/04-04-24-14-52-45/json/bz_128_noise_samples_8_sample_weights_1_dataset_names_droid_local_cams_2cams_goal_mode_None_truncated_geom_factor_0.3_ldkeys_proprio-lang_visenc_VisualCore_fuser_None.json

Which Python code command should I execute?

[ashwin-balakrishna96]

The first command corresponds to training on droid_100, which is a small subset of DROID that could be useful if you want to do a quick sanity check (see the droid_100 identifier in the generated command). The second command corresponds to training on all of DROID (see the droid_local identified in the generated command).

[ashwin-balakrishna96]

You can see the robomimic documentation for more information on how the config_gen files work more generally, but the idea is that you can define a number of different hyperparameters that you want to sweep over in robomimic/scripts/config_gen/droid_runs_language_conditioned_rlds.py and then when you run this script, it will generate commands corresponding to every experiment you need to run.

[jileimao]

You can see the robomimic documentation for more information on how the config_gen files work more generally, but the idea is that you can define a number of different hyperparameters that you want to sweep over in robomimic/scripts/config_gen/droid_runs_language_conditioned_rlds.py and then when you run this script, it will generate commands corresponding to every experiment you need to run.

hi, I add a co-data in /robomimic/scripts/config_gen/droid_runs_language_conditioned_rlds.py, then I get two exps commands. Am I supposed to run the two commands one by one? or what should I do then?