config.yaml

resume: false
device: cuda
use_amp: true
seed: 100000
dataset_repo_id: lerobot/pusht
video_backend: pyav
training:
  offline_steps: 1000000
  num_workers: 24
  batch_size: 24
  eval_freq: 10000
  log_freq: 1000
  save_checkpoint: true
  save_freq: 50000
  online_steps: 0
  online_rollout_n_episodes: 1
  online_rollout_batch_size: 1
  online_steps_between_rollouts: 1
  online_sampling_ratio: 0.5
  online_env_seed: null
  online_buffer_capacity: null
  online_buffer_seed_size: 0
  do_online_rollout_async: false
  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
  save_model: true
  grad_clip_norm: 50
  lr: 0.0001
  min_lr: 0.0001
  lr_cycle_steps: 300000
  weight_decay: 1.0e-05
  delta_timestamps:
    observation.image:
    - -1.5
    - -1.4
    - -1.3
    - -1.2
    - -1.1
    - -1.0
    - -0.9
    - -0.8
    - -0.7
    - -0.6
    - -0.5
    - -0.1
    - 0.0
    observation.state:
    - -1.5
    - -1.4
    - -1.3
    - -1.2
    - -1.1
    - -1.0
    - -0.9
    - -0.8
    - -0.7
    - -0.6
    - -0.5
    - -0.1
    - 0.0
    action:
    - -1.5
    - -1.4
    - -1.3
    - -1.2
    - -1.1
    - -1.0
    - -0.9
    - -0.8
    - -0.7
    - -0.6
    - -0.5
    - -0.1
    - 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
    - 1.6
    - 1.7
    - 1.8
    - 1.9
eval:
  n_episodes: 100
  batch_size: 100
  use_async_envs: false
wandb:
  enable: true
  disable_artifact: false
  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: pixels_agent_pos
    render_mode: rgb_array
    visualization_width: 384
    visualization_height: 384
override_dataset_stats:
  observation.image:
    mean:
    - - - 0.485
    - - - 0.456
    - - - 0.406
    std:
    - - - 0.229
    - - - 0.224
    - - - 0.225
  observation.state:
    min:
    - 0.0
    - 0.0
    max:
    - 512.0
    - 512.0
  action:
    min:
    - 0.0
    - 0.0
    max:
    - 512.0
    - 512.0
policy:
  name: dot
  n_obs_steps: 3
  train_horizon: 20
  inference_horizon: 20
  lookback_obs_steps: 10
  lookback_aug: 5
  input_shapes:
    observation.image:
    - 3
    - 96
    - 96
    observation.state:
    - ${env.state_dim}
  output_shapes:
    action:
    - ${env.action_dim}
  input_normalization_modes:
    observation.image: mean_std
    observation.state: min_max
  output_normalization_modes:
    action: min_max
  vision_backbone: resnet18
  pretrained_backbone_weights: ResNet18_Weights.IMAGENET1K_V1
  rescale_shape:
  - 96
  - 96
  lora_rank: 20
  merge_lora: true
  crop_scale: 0.8
  state_noise: 0.01
  noise_decay: 0.999995
  pre_norm: true
  dim_model: 128
  n_heads: 8
  dim_feedforward: 512
  n_decoder_layers: 8
  dropout: 0.1
  alpha: 0.75
  train_alpha: 0.9
  predict_every_n: 1
  return_every_n: 2