resume: false
device: cuda
use_amp: true
seed: 100000
dataset_repo_id: lerobot/pusht_keypoints
video_backend: pyav
training:
  offline_steps: 100000
  online_steps: 0
  online_steps_between_rollouts: ???
  online_sampling_ratio: 0.5
  online_env_seed: ???
  eval_freq: 10000
  log_freq: 250
  save_checkpoint: true
  save_freq: 25000
  num_workers: 8
  batch_size: 64
  image_transforms:
    enable: false
    max_num_transforms: 3
    random_order: false
    brightness:
      weight: 1
      min_max:
      - 0.8
      - 1.2
    contrast:
      weight: 1
      min_max:
      - 0.8
      - 1.2
    saturation:
      weight: 1
      min_max:
      - 0.5
      - 1.5
    hue:
      weight: 1
      min_max:
      - -0.05
      - 0.05
    sharpness:
      weight: 1
      min_max:
      - 0.8
      - 1.2
  lr: 3.0e-05
  lr_backbone: 1.0e-05
  weight_decay: 0.0001
  grad_clip_norm: 10
  delta_timestamps:
    action:
    - 0.0
    - 0.1
    - 0.2
    - 0.3
    - 0.4
    - 0.5
    - 0.6
    - 0.7
    - 0.8
    - 0.9
    - 1.0
    - 1.1
    - 1.2
    - 1.3
    - 1.4
    - 1.5
eval:
  n_episodes: 50
  batch_size: 50
  use_async_envs: false
wandb:
  enable: true
  disable_artifact: true
  project: lerobot
  notes: ''
fps: 10
env:
  name: pusht
  task: PushT-v0
  image_size: 96
  state_dim: 2
  action_dim: 2
  fps: ${fps}
  episode_length: 300
  gym:
    obs_type: environment_state_agent_pos
    render_mode: rgb_array
    visualization_width: 384
    visualization_height: 384
policy:
  name: act
  n_obs_steps: 1
  chunk_size: 16
  n_action_steps: 16
  input_shapes:
    observation.environment_state:
    - 16
    observation.state:
    - ${env.state_dim}
  output_shapes:
    action:
    - ${env.action_dim}
  input_normalization_modes:
    observation.environment_state: min_max
    observation.state: min_max
  output_normalization_modes:
    action: min_max
  vision_backbone: resnet18
  pretrained_backbone_weights: ResNet18_Weights.IMAGENET1K_V1
  replace_final_stride_with_dilation: false
  pre_norm: false
  dim_model: 512
  n_heads: 8
  dim_feedforward: 3200
  feedforward_activation: relu
  n_encoder_layers: 4
  n_decoder_layers: 4
  use_vae: true
  latent_dim: 32
  n_vae_encoder_layers: 4
  temporal_ensemble_momentum: null
  dropout: 0.1
  kl_weight: 10.0