runner:
    class_name: OnPolicyRunner
    # General
    num_steps_per_env: 24 # Number of steps per environment per iteration
    max_iterations: 1500 # Number of policy updates
    seed: 1
    # Observations
    obs_groups: { "actor": ["policy"], "critic": ["policy", "privileged"] } # Maps from observation sets to groups. See `vec_env.py` for more information
    # Logging parameters
    save_interval: 50 # Check for potential saves every `save_interval` iterations
    experiment_name: walking_experiment
    run_name: ""
    # Logging writer
    logger: tensorboard # tensorboard, neptune, wandb
    neptune_project: rsl_rl
    wandb_project: rsl_rl

    # Actor
    actor:
        class_name: MLPModel
        hidden_dims: [256, 256, 256]
        activation: elu
        obs_normalization: false
        stochastic: true
        init_noise_std: 1.0
        noise_std_type: "scalar" # 'scalar' or 'log'
        state_dependent_std: false

    # Critic
    critic:
        class_name: MLPModel
        hidden_dims: [256, 256, 256]
        activation: elu
        obs_normalization: false
        stochastic: false

    # Algorithm
    algorithm:
        class_name: PPO
        # Training
        optimizer: adam # adam, adamw, sgd, rmsprop
        learning_rate: 0.001
        num_learning_epochs: 5
        num_mini_batches: 4 # mini batch size = num_envs * num_steps / num_mini_batches
        schedule: adaptive # adaptive, fixed
        # Value function
        value_loss_coef: 1.0
        clip_param: 0.2
        use_clipped_value_loss: true
        # Surrogate loss
        desired_kl: 0.01
        entropy_coef: 0.01
        gamma: 0.99
        lam: 0.95
        max_grad_norm: 1.0
        # Miscellaneous
        normalize_advantage_per_mini_batch: false

        # Random network distillation
        rnd_cfg:
            weight: 0.0 # Initial weight of the RND reward
            weight_schedule: null # This is a dictionary with a required key called "mode". Please check the RND module for more information
            reward_normalization: false # Whether to normalize RND reward
            # Learning parameters
            learning_rate: 0.001 # Learning rate for RND
            # Network parameters
            num_outputs: 1 # Number of outputs of RND network. Note: if -1, then the network will use dimensions of the observation
            predictor_hidden_dims: [-1] # Hidden dimensions of predictor network
            target_hidden_dims: [-1] # Hidden dimensions of target network

        # Symmetry augmentation
        symmetry_cfg:
            use_data_augmentation: true # This adds symmetric trajectories to the batch
            use_mirror_loss: false # This adds symmetry loss term to the loss function
            data_augmentation_func: null # String containing the module and function name to import
            # Example: "legged_gym.envs.locomotion.anymal_c.symmetry:get_symmetric_states"
            #
            # .. code-block:: python
            #
            #     @torch.no_grad()
            #     def get_symmetric_states(
            #        env: VecEnv, obs: Optional[torch.Tensor] = None, actions: Optional[torch.Tensor] = None,
            #     ) -> Tuple[torch.Tensor, torch.Tensor]:
            #
            mirror_loss_coeff: 0.0 # Coefficient for symmetry loss term. If 0, no symmetry loss is used