| seed: 1 |
| device: cuda:0 |
| num_steps_per_env: 24 |
| max_iterations: 1000000 |
| empirical_normalization: true |
| policy: |
| class_name: ActorCriticMNMLP |
| init_noise_std: 1.0 |
| noise_std_type: scalar |
| actor_hidden_dims: |
| - 2048 |
| - 1024 |
| - 512 |
| critic_hidden_dims: |
| - 2048 |
| - 1024 |
| - 512 |
| activation: elu |
| algorithm: |
| class_name: PPO_MNMLP |
| num_learning_epochs: 5 |
| num_mini_batches: 4 |
| learning_rate: 0.001 |
| schedule: adaptive |
| gamma: 0.99 |
| lam: 0.95 |
| entropy_coef: 0.005 |
| desired_kl: 0.01 |
| max_grad_norm: 1.0 |
| value_loss_coef: 1.0 |
| use_clipped_value_loss: true |
| clip_param: 0.2 |
| normalize_advantage_per_mini_batch: false |
| symmetry_cfg: null |
| rnd_cfg: null |
| clip_actions: null |
| save_interval: 500 |
| experiment_name: DataARCRL-1108-train |
| run_name: safe-DataARCRL-1108-train |
| logger: tensorboard |
| neptune_project: isaaclab |
| wandb_project: isaaclab |
| resume: false |
| load_run: .* |
| load_checkpoint: model_.*.pt |
|
|
