draccus parsing issue

[mqp2259]

We encountered this issue when we tried to evaluate a trained policy.

Anyone can help with it?

[shantanuparab-tr]

In the same location as the policy.path in your case outputs/train/act_task_pickup/checkpoints/000200/pretrained_model you will have a file config.json. Currently this file might look like this

    "n_obs_steps": 1,
    "normalization_mapping": {
        "VISUAL": "MEAN_STD",
        "STATE": "MEAN_STD",
        "ACTION": "MEAN_STD"
    },

The key-value pair type: act will be missing. If you manually add this entry you should be able to fix the above problem.

    "type": "act",
    "n_obs_steps": 1,
    "normalization_mapping": {
        "VISUAL": "MEAN_STD",
        "STATE": "MEAN_STD",
        "ACTION": "MEAN_STD"
    },

We will be coming up with a solution for this soon. This is a quick fix for now.