AllegroHandLSTM_BigPPO.yaml

params:
  seed: ${...seed}

  algo:
    name: a2c_continuous

  model:
    name: continuous_a2c_logstd

  network:
    name: complex_net
    separate: False
    space:
      continuous:
        mu_activation: None
        sigma_activation: None
        mu_init:
          name: default
        sigma_init:
          name: const_initializer
          val: 0
        fixed_sigma: True
      
    inputs:
      dof_pos_offset_randomized: {}
      object_pose_delayed_randomized: {}
      goal_pose: {}
      goal_relative_rot_delayed_randomized: {}
      last_actions: {}

    mlp:
      units: [512]
      activation: elu
      d2rl: False

      initializer:
        name: default
      regularizer:
        name: None
    rnn:
        name: lstm
        units: 1024
        layers: 1
        before_mlp: True
        layer_norm: True

  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: ${resolve_default:AllegroHandAsymmLSTM,${....experiment}}
    full_experiment_name: ${.name}
    env_name: rlgpu
    multi_gpu: ${....multi_gpu}
    ppo: True
    mixed_precision: False
    normalize_input: True
    normalize_value: True
    use_smooth_clamp: True
    num_actors: ${....task.env.numEnvs}
    reward_shaper:
      scale_value: 1.0
    normalize_advantage: True
    gamma: 0.998
    tau: 0.95
    learning_rate: 1e-4
    lr_schedule: adaptive
    schedule_type: standard
    kl_threshold: 0.016
    score_to_win: 100000
    max_epochs: ${resolve_default:1000000,${....max_iterations}}
    save_best_after: 200
    save_frequency: 500
    print_stats: True
    grad_norm: 1.0
    entropy_coef: 0.0
    truncate_grads: True
    e_clip: 0.2
    horizon_length: 16
    minibatch_size: 16384
    mini_epochs: 4
    critic_coef: 4
    clip_value: True
    seq_length: 16
    bounds_loss_coef: 0.0001

    central_value_config:
      minibatch_size: 16384 #32768
      mini_epochs: 4
      learning_rate: 1e-4
      kl_threshold: 0.016
      clip_value: True
      normalize_input: True
      truncate_grads: True
      use_smooth_clamp: True

      network:
        name: complex_net
        central_value: True
        inputs:
          dof_pos: {}
          dof_pos_offset_randomized: {}
          dof_vel: {}
          dof_torque: {}
          object_pose: {}
          object_vels: {}
          goal_pose: {}
          goal_relative_rot: {}
          object_pose_delayed_randomized: {}
          goal_relative_rot_delayed_randomized: {}
          object_obs_delayed_age: {}
          # ft_states: {}
          # ft_force_torques: {}
          last_actions: {}
        mlp:
          units: [512, 512, 256, 128] #[256] # 
          activation: elu
          d2rl: False
          initializer:
            name: default
          regularizer:
            name: None
        # rnn:
        #   name: lstm
        #   units: 512
        #   layers: 1
        #   before_mlp: True
        #   layer_norm: True

    player:
      deterministic: True
      use_vecenv: True
      games_num: 1000000
      print_stats: False