× type: MemoryGym
name: Endless-MysteryPath-v0
frame_skip: 1
last_action_to_obs: False
last_reward_to_obs: False
obs_stacks: 1
grayscale: False
resize_vis_obs: [84, 84]
positional_encoding: False
reset_params:
        start-seed: 200002
        num-seeds: 1
        max_steps: -1
        agent_scale: 0.25
        agent_speed: 3.0
        show_origin: True
        show_past_path: True
        show_background: False
        show_stamina: False
        visual_feedback: True
        camera_offset_scale: 5.0
        stamina_level: 20
        reward_fall_off: 0.0
        reward_path_progress: 0.1
        reward_step: 0.0
        seed: 200002
reward_normalization: 0
positional_endocing: False
× load_model: False
model_path: ./checkpoints/mg_17r/20220112-065616_2/mg_17r-208.pt
checkpoint_interval: 500
activation: relu
vis_encoder: cnn
vec_encoder: linear
num_vec_encoder_units: 128
hidden_layer: default
num_hidden_layers: 1
num_hidden_units: 512
vec_encoder_layer: linear
recurrence:
        layer_type: gru
        sequence_length: -1
        hidden_state_size: 512
        hidden_state_init: zero
        reset_hidden_state: True
        residual: False
        num_layers: 1
× n_workers: 32
worker_steps: 512
× algorithm: PPO
resume_at: 0
gamma: 0.995
lamda: 0.95
updates: 50000
epochs: 3
n_mini_batches: 8
advantage_normalization: no
value_coefficient: 0.5
max_grad_norm: 0.25
learning_rate_schedule:
        initial: 0.000275
        final: 1e-05
        power: 1.0
        max_decay_steps: 10000
beta_schedule:
        initial: 0.0001
        final: 1e-06
        power: 1.0
        max_decay_steps: 10000
clip_range_schedule:
        initial: 0.1
        final: 0.1
        power: 1.0
        max_decay_steps: 10000
obs_reconstruction_schedule: {'initial': 0.0}
ground_truth_estimator_schedule: {'initial': 0.0}
refresh_buffer_epoch: -1
share_parameters: True

Endless Mystery Path
+ Gated Recurrent Unit