AnymalTerrainPPO.yaml

params:
  seed: ${...seed}

  algo:
    name: a2c_continuous

  model:
    name: continuous_a2c_logstd

  network:
    name: actor_critic
    separate: True

    space:
      continuous:
        mu_activation: None
        sigma_activation: None
        mu_init:
          name: default
        sigma_init:
          name: const_initializer
          val: 0. # std = 1.
        fixed_sigma: True

    mlp:
      units: [512, 256, 128]
      activation: elu
      d2rl: False

      initializer:
        name: default
      regularizer:
        name: None
    # rnn:
    #   name: lstm
    #   units: 128
    #   layers: 1
    #   before_mlp: True
    #   concat_input: True
    #   layer_norm: False


  load_checkpoint: ${if:${...checkpoint},True,False} # flag which sets whether to load the checkpoint
  load_path: ${...checkpoint} # path to the checkpoint to load

  config:
    name: ${resolve_default:AnymalTerrain,${....experiment}}
    full_experiment_name: ${.name}
    device: ${....rl_device}
    device_name: ${....rl_device}
    env_name: rlgpu
    multi_gpu: ${....multi_gpu}
    ppo: True
    mixed_precision: False # True
    normalize_input: True
    normalize_value: True
    normalize_advantage: True
    value_bootstrap: True
    clip_actions: False
    num_actors: ${....task.env.numEnvs}
    reward_shaper:
      scale_value: 1.0
    gamma: 0.99
    tau: 0.95
    e_clip: 0.2
    entropy_coef: 0.001
    learning_rate: 3.e-4 # overwritten by adaptive lr_schedule
    lr_schedule: adaptive
    kl_threshold: 0.008 # target kl for adaptive lr
    truncate_grads: True
    grad_norm: 1.
    horizon_length: 48
    minibatch_size: 16384
    mini_epochs: 5
    critic_coef: 2
    clip_value: True
    seq_length: 4 # only for rnn
    bounds_loss_coef: 0.

    max_epochs: ${resolve_default:2000,${....max_iterations}}
    save_best_after: 100
    score_to_win: 20000
    save_frequency: 50
    print_stats: True