seed: 42
num_steps_per_env: 40
max_iterations: 7500
obs_groups:
  actor: !!python/tuple
  - policy
  critic: !!python/tuple
  - critic
save_interval: 50
experiment_name: cyclops_finger_reorient
run_name: slurm_29095053_resumed
logger: wandb
wandb_project: wuji_reorient_mjlab
wandb_tags: !!python/tuple []
resume: true
load_run: 2026-07-25_11-46-06_slurm_29033121
load_checkpoint: model_3150.pt
clip_actions: null
upload_model: true
class_name: OnPolicyRunner
actor:
  hidden_dims: !!python/tuple
  - 512
  - 256
  - 128
  activation: elu
  obs_normalization: true
  cnn_cfg: null
  distribution_cfg:
    class_name: SoftplusGaussianDistribution
    init_std: 0.5
    min_std: 0.2
  rnn_type: null
  rnn_hidden_dim: 256
  rnn_num_layers: 1
  class_name: MLPModel
critic:
  hidden_dims: !!python/tuple
  - 512
  - 512
  - 256
  - 128
  activation: elu
  obs_normalization: true
  cnn_cfg: null
  distribution_cfg: null
  rnn_type: null
  rnn_hidden_dim: 256
  rnn_num_layers: 1
  class_name: MLPModel
algorithm:
  num_learning_epochs: 4
  num_mini_batches: 32
  learning_rate: 0.0001
  schedule: fixed
  gamma: 0.99
  lam: 0.95
  entropy_coef: 0.001
  desired_kl: 0.01
  max_grad_norm: 1.0
  value_loss_coef: 0.5
  use_clipped_value_loss: false
  clip_param: 0.2
  normalize_advantage_per_mini_batch: false
  optimizer: adam
  share_cnn_encoders: false
  class_name: PPO