{
    "type": "act",
    "n_obs_steps": 1,
    "normalization_mapping": {
        "VISUAL": "MEAN_STD",
        "STATE": "MEAN_STD",
        "ACTION": "MEAN_STD",
        "ENV": "MEAN_STD"
    },
    "input_features": {
        "observation.right.joint": {
            "type": "STATE",
            "shape": [
                8
            ]
        },
        "observation.environment.tblock.center.xy": {
            "type": "ENV",
            "shape": [
                2
            ]
        },
        "observation.environment.tblock.center.xaxis": {
            "type": "ENV",
            "shape": [
                3
            ]
        }
    },
    "output_features": {
        "action.right.joint": {
            "type": "ACTION",
            "shape": [
                8
            ]
        }
    },
    "device": "cuda",
    "use_amp": false,
    "chunk_size": 100,
    "n_action_steps": 10,
    "vision_backbone": "resnet18",
    "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
    "replace_final_stride_with_dilation": false,
    "use_segmentation_backbone": false,
    "segmentation_backbone_type": "robot",
    "segmentation_backbone_sam_version": "michaelyuanqwq/roboengine-sam",
    "segmentation_backbone_tokenizer_version": "YxZhang/evf-sam2-multitask",
    "segmentation_backbone_image_size": 224,
    "segmentation_backbone_device": "cuda",
    "segmentation_backbone_instruction": "robot",
    "segmentation_backbone_feature_dim": 512,
    "pre_norm": false,
    "dim_model": 512,
    "n_heads": 8,
    "dim_feedforward": 3200,
    "feedforward_activation": "relu",
    "n_encoder_layers": 4,
    "n_decoder_layers": 1,
    "use_vae": true,
    "latent_dim": 32,
    "n_vae_encoder_layers": 4,
    "temporal_ensemble_coeff": null,
    "dropout": 0.1,
    "kl_weight": 10.0,
    "optimizer_lr": 1e-05,
    "optimizer_weight_decay": 0.0001,
    "optimizer_lr_backbone": 1e-05,
    "freeze_vision_backbone": false,
    "use_prediction_head": false,
    "prediction_head_type": "regression",
    "prediction_head_output_dim": [
        1
    ],
    "prediction_head_hidden_dim": 256,
    "prediction_head_dropout": 0.1,
    "prediction_head_loss_weight": [
        1.0
    ],
    "prediction_target_key": [
        "prediction_target"
    ]
}