config.json

{
  "advantage_scaling": 3.0,
  "cem_iterations": 6,
  "consistency_coeff": 20.0,
  "discount": 0.98,
  "elite_weighting_temperature": 0.5,
  "expectile_weight": 0.9,
  "gaussian_mean_momentum": 0.1,
  "horizon": 5,
  "image_encoder_hidden_dim": 32,
  "input_normalization_modes": {
    "observation.environment_state": "min_max",
    "observation.state": "min_max"
  },
  "input_shapes": {
    "observation.environment_state": [
      16
    ],
    "observation.state": [
      2
    ]
  },
  "latent_dim": 50,
  "max_random_shift_ratio": 0.0476,
  "max_std": 2.0,
  "min_std": 0.05,
  "mlp_dim": 512,
  "n_action_repeats": 1,
  "n_action_steps": 5,
  "n_elites": 50,
  "n_gaussian_samples": 512,
  "n_pi_samples": 51,
  "output_normalization_modes": {
    "action": "min_max"
  },
  "output_shapes": {
    "action": [
      2
    ]
  },
  "pi_coeff": 0.5,
  "q_ensemble_size": 5,
  "reward_coeff": 0.5,
  "state_encoder_hidden_dim": 256,
  "target_model_momentum": 0.995,
  "temporal_decay_coeff": 0.5,
  "uncertainty_regularizer_coeff": 1.0,
  "use_mpc": true,
  "value_coeff": 0.1
}