Deployed f569f4c with MkDocs version: 1.6.1

configs/architecture/centralized.yaml

@@ -0,0 +1,22 @@
+# Centralized Actor-Critic Architecture
+# Baseline configuration with a single global sensor and motor.
+
+# Default values are defined in CentralizedConfig dataclass.
+# Use this configuration for standard PPO experiments.
+
+name: "centralized"
+sensor:
+  hidden_dims: [300, 300, 300]
+  activation: "tanh"
+
+motor:
+  hidden_dims: []
+  activation: "tanh"
+
+feature_extractor:
+  hidden_dims: [300, 300, 300]
+  activation: "tanh"
+
+critic:
+  hidden_dims: []
+  activation: "tanh"

configs/architecture/decentralized.yaml

@@ -0,0 +1,32 @@
+# Decentralized Actor Architecture (NerveNet-MLP variant)
+# Multi-agent/distributed configuration using local sensors, propagators, and motors.
+
+# Default values are defined in DecentralizedConfig dataclass.
+# Use this configuration for decentralized execution experiments.
+
+name: "decentralized"
+sensor:
+  hidden_dims: [300, 300, 300]
+  activation: "tanh"
+
+propagator:
+  hidden_dims: [300, 300, 300]
+  activation: "tanh"
+
+motor:
+  hidden_dims: []
+  activation: "tanh"
+
+feature_extractor:
+  hidden_dims: [300, 300, 300]
+  activation: "tanh"
+
+critic:
+  hidden_dims: []
+  activation: "tanh"
+
+# Synchronous message-passing rounds per control step
+message_passing_steps: 4
+
+# Connectivity topology (e.g., ring, fully_connected)
+topology_type: "fully_connected"

configs/arena/default.yaml

@@ -0,0 +1,8 @@
+# Default Arena Configuration
+# Base aquarium environment settings.
+
+size: [10.0, 5.0]
+sand_ground_color: true
+attach_target: true
+wall_height: 1.5
+wall_thickness: 0.1

configs/centralized-final.yaml

@@ -0,0 +1,71 @@
+# Custom Main Configuration
+#
+# Use with:
+#   uv run python scripts/train.py --config-name main_config_custom
+#
+# This keeps the project defaults intact while giving you a single custom
+# training entrypoint you can edit freely.
+
+defaults:
+  - brittle_star_config
+  - experiment: base
+  - logging: default
+  - evaluation: default
+  - ppo: default
+  - architecture: centralized
+  - morphology: 5_arms_full
+  - arena: default
+  - environment: directed_locomotion
+  - simulation: default
+  - _self_
+
+morphology:
+  morph_mode: CENTRALIZED
+
+experiment:
+  exp_name: "final-models/centralized/"
+  seed: 42
+  torch_deterministic: true
+  cuda: true
+
+logging:
+  track: true
+  save_model: true
+  save_checkpoints: true
+  upload_final_model: true
+  upload_checkpoints: true
+  checkpoint_frequency: 20
+  wandb_project_name: "final-models"
+
+evaluation:
+  evaluate_checkpoints: true
+  eval_max_steps: 2000
+  eval_seed: 0
+
+ppo:
+  learning_rate: 0.0001
+  total_timesteps: 16384000
+  num_envs: 128
+  num_steps: 64
+  anneal_lr: true
+  gamma: 0.99
+  gae_lambda: 0.95
+  num_minibatches: 32
+  update_epochs: 4
+  norm_adv: true
+  clip_coef: 0.2
+  clip_vloss: true
+  ent_coef: 0.001
+  vf_coef: 1.0
+  max_grad_norm: 0.5
+  target_kl: 0.02
+
+environment:
+  simulation_time: 100000.0
+  target_distance: 3.0
+
+hydra:
+  job:
+    chdir: true
+  run:
+    dir: ${experiment.base_run_dir}/${experiment.exp_name}/${now:%Y-%m-%d}/${now:%H-%M-%S}

configs/environment/dir_loc_further.yaml

@@ -0,0 +1,2 @@
+simulation_time: 50000.0
+target_distance: 3.0

configs/environment/directed_locomotion.yaml

@@ -0,0 +1,12 @@
+# Directed Locomotion Environment
+# Baseline task setting.
+
+task: DIRECTED_LOCOMOTION
+simulation_time: 100000.0
+num_physics_steps_per_control_step: 10
+time_scale: 2
+camera_ids: [0, 1]
+render_size: [480, 640]
+joint_randomization_noise_scale: 0.0
+target_distance: 3.0
+light_perlin_noise_scale: 0

configs/environment/light_escape.yaml

@@ -0,0 +1,12 @@
+# Light Escape Environment
+# Advanced task requiring movement away from light source.
+
+task: LIGHT_ESCAPE
+simulation_time: 100000.0
+num_physics_steps_per_control_step: 10
+time_scale: 2
+camera_ids: [0, 1]
+render_size: [480, 640]
+joint_randomization_noise_scale: 0.0
+target_distance: 3.0
+light_perlin_noise_scale: 200 # Must be integer factor of 200

configs/evaluation/default.yaml

@@ -0,0 +1,8 @@
+# Default Evaluation Configuration
+# Settings used for checkpoint evaluation during training.
+
+evaluate_checkpoints: false
+# Max number of control steps during evaluation rollout.
+eval_max_steps: 2000
+# Seed for deterministic evaluation reset.
+eval_seed: 0

configs/evaluation/poster.yaml

@@ -0,0 +1,23 @@
+# @package evaluation
+# Configuration for the models used in the poster comparison.
+
+# Standard evaluation settings
+evaluate_checkpoints: false
+eval_max_steps: 5000
+eval_seed: 0
+
+# Cross-model comparison settings
+# We use 10 episodes to get a more robust average for the final poster results.
+comparison_base_seed: 0
+comparison_num_episodes: 2
+comparison_output_csv: "runs/evaluation/comparison.csv"
+
+# Paths to the .cleanrl_model files to be compared (relative to workspace root).
+comparison_models:
+  - "runs/input-space-2-arms/2026-05-02/08-14-58/final_model.flax"
+
+# Path to the morphologies to evaluate against.
+comparison_morphologies:
+  - "configs/morphology/5_arms_full.yaml"
+  - "configs/morphology/3_arms.yaml"
+  - "configs/morphology/2_arms.yaml"

configs/experiment/base.yaml

@@ -0,0 +1,7 @@
+# Base Experiment Configuration
+# Default values align with ExperimentConfig dataclass.
+
+exp_name: "brittle_star_ppo"
+seed: 1
+torch_deterministic: true
+cuda: true

configs/experiment/dev_test.yaml

@@ -0,0 +1,7 @@
+# Testing Experiment Configuration
+# Quick experiment for local development/testing.
+
+exp_name: "dev_test_brittle_star"
+seed: 42
+torch_deterministic: true
+cuda: true

configs/experiment/hpc_smoke_test.yaml

@@ -0,0 +1,7 @@
+# HPC Smoke Test Configuration
+# Uses minimal settings but simulates HPC environment.
+
+exp_name: "hpc_smoke_test"
+seed: 123
+torch_deterministic: true
+cuda: true

configs/experiment/long_2arm.yaml

@@ -0,0 +1,6 @@
+# Testing chicken dinner 4 but further distance.
+
+exp_name: "long2arm"
+seed: 123
+torch_deterministic: true
+cuda: true

configs/fully-connected-final.yaml

@@ -0,0 +1,74 @@
+# Custom Main Configuration
+#
+# Use with:
+#   uv run python scripts/train.py --config-name main_config_custom
+#
+# This keeps the project defaults intact while giving you a single custom
+# training entrypoint you can edit freely.
+
+defaults:
+  - brittle_star_config
+  - experiment: base
+  - logging: default
+  - evaluation: default
+  - ppo: default
+  - architecture: decentralized
+  - morphology: 5_arms_full
+  - arena: default
+  - environment: directed_locomotion
+  - simulation: default
+  - _self_
+
+architecture:
+  topology_type: "fully_connected"
+
+morphology:
+  morph_mode: FULLY_CONNECTED
+
+experiment:
+  exp_name: "final-models/fully-connected/"
+  seed: 42
+  torch_deterministic: true
+  cuda: true
+
+logging:
+  track: true
+  save_model: true
+  save_checkpoints: true
+  upload_final_model: true
+  upload_checkpoints: true
+  checkpoint_frequency: 20
+  wandb_project_name: "final-models"
+
+evaluation:
+  evaluate_checkpoints: true
+  eval_max_steps: 2000
+  eval_seed: 0
+
+ppo:
+  learning_rate: 0.0001
+  total_timesteps: 16384000
+  num_envs: 128
+  num_steps: 64
+  anneal_lr: true
+  gamma: 0.99
+  gae_lambda: 0.95
+  num_minibatches: 32
+  update_epochs: 4
+  norm_adv: true
+  clip_coef: 0.2
+  clip_vloss: true
+  ent_coef: 0.001
+  vf_coef: 1.0
+  max_grad_norm: 0.5
+  target_kl: 0.02
+
+environment:
+  simulation_time: 100000.0
+  target_distance: 3.0
+
+hydra:
+  job:
+    chdir: true
+  run:
+    dir: ${experiment.base_run_dir}/${experiment.exp_name}/${now:%Y-%m-%d}/${now:%H-%M-%S}

configs/logging/default.yaml

@@ -0,0 +1,13 @@
+# Default Logging Configuration
+# Offline local-only setup (WandB disabled).
+
+track: false
+wandb_project_name: "PPO-Modularity"
+wandb_entity: "SEL3-2026-Groep-4"
+capture_video: false
+save_model: true
+save_checkpoints: true
+checkpoint_frequency: 100
+upload_final_model: false
+upload_checkpoints: false
+hf_entity: ""

configs/logging/hpc.yaml

@@ -0,0 +1,9 @@
+track: true
+wandb_project_name: "hpc-default"
+wandb_entity: "SEL3-2026-Groep-4"
+save_model: true
+save_checkpoints: true
+upload_final_model: true
+upload_checkpoints: true
+checkpoint_frequency: 100
+hf_entity: ""

configs/logging/wandb_enabled.yaml

@@ -0,0 +1,13 @@
+# WandB Enabled Logging Configuration
+# For production/cloud experiments with weights synced.
+
+track: true
+wandb_project_name: "default-project"
+wandb_entity: "SEL3-2026-Groep-4"
+capture_video: false
+save_model: true
+save_checkpoints: true
+checkpoint_frequency: 100
+upload_final_model: true
+upload_checkpoints: false
+hf_entity: ""

configs/main_config.yaml

@@ -0,0 +1,23 @@
+# Brittle Star Project - Main Configuration
+# This file defines the default composition of the hierarchical configuration.
+# Sub-configs are loaded from the relative directories.
+
+defaults:
+  - brittle_star_config
+  - experiment: base
+  - logging: default
+  - evaluation: default
+  - ppo: default
+  - architecture: centralized
+  - morphology: 5_arms_full
+  - arena: default
+  - environment: directed_locomotion
+  - obs_bounds: default
+  - simulation: default
+  - _self_
+
+hydra:
+  job:
+    chdir: True
+  run:
+    dir: ${experiment.base_run_dir}/${experiment.exp_name}/${now:%Y-%m-%d}/${now:%H-%M-%S}

configs/morphology/2_arms.yaml

@@ -0,0 +1,5 @@
+# 2 Arms Morphology Configuration
+
+segments_per_arm: [4, 0, 4, 0, 0]
+use_p_control: true
+use_torque_control: false

configs/morphology/2_arms_decentralized.yaml

@@ -0,0 +1,6 @@
+# 2 Arms Morphology Configuration
+
+segments_per_arm: [4, 0, 4, 0, 0]
+use_p_control: true
+use_torque_control: false
+morph_mode: FULLY_CONNECTED

configs/morphology/3_arms.yaml

@@ -0,0 +1,6 @@
+# 3 Arms Morphology Configuration
+# Symmetric amputation (arms 1 and 3 removed).
+
+segments_per_arm: [4, 0, 4, 0, 4]
+use_p_control: true
+use_torque_control: false

configs/morphology/5_arms_damaged.yaml

@@ -0,0 +1,6 @@
+# 5 Arms Full Morphology Configuration
+# Baseline 5-arm brittle star.
+
+segments_per_arm: [4, 4, 0, 4, 4]
+use_p_control: true
+use_torque_control: false

configs/morphology/5_arms_full.yaml

@@ -0,0 +1,6 @@
+# 5 Arms Full Morphology Configuration
+# Baseline 5-arm brittle star.
+
+segments_per_arm: [4, 4, 4, 4, 4]
+use_p_control: true
+use_torque_control: false

configs/morphology/5_arms_full_fullconnected.yaml

@@ -0,0 +1,7 @@
+# 5 Arms Full Morphology Configuration
+# Baseline 5-arm brittle star.
+
+segments_per_arm: [4, 4, 4, 4, 4]
+use_p_control: true
+use_torque_control: false
+morph_mode: FULLY_CONNECTED

configs/morphology/partial_amputation.yaml

@@ -0,0 +1,6 @@
+# Partial Amputation Configuration
+# Random partial amputation for robustness testing.
+
+segments_per_arm: [4, 2, 4, 4, 4]
+use_p_control: true
+use_torque_control: false

configs/obs_bounds/default.yaml

@@ -0,0 +1 @@
+# Defaults provided by dataclass

configs/ppo/chickendinnerwinner.yaml

@@ -0,0 +1,16 @@
+anneal_lr: true
+clip_coef: 0.2
+clip_vloss: true
+ent_coef: 0.001
+gae_lambda: 0.95
+gamma: 0.99
+learning_rate: 0.0001
+max_grad_norm: 0.5
+norm_adv: true
+num_envs: 32
+num_minibatches: 32
+num_steps: 64
+target_kl: 0.02
+total_timesteps: 12288000
+update_epochs: 4
+vf_coef: 1.0

configs/ppo/debug.yaml

@@ -0,0 +1,16 @@
+learning_rate: 0.0003
+total_timesteps: 409600
+num_envs: 32
+num_steps: 32
+anneal_lr: true
+gamma: 0.99
+gae_lambda: 0.95
+num_minibatches: 32
+update_epochs: 4
+norm_adv: true
+clip_coef: 0.2
+clip_vloss: true
+ent_coef: 0.005
+vf_coef: 1.0
+max_grad_norm: 0.5
+target_kl: null

configs/ppo/default.yaml

@@ -0,0 +1,19 @@
+# Default PPO Configuration
+# Standard hyperparams from original codebase.
+
+learning_rate: 0.00025
+total_timesteps: 10000000
+num_envs: 100
+num_steps: 128
+anneal_lr: true
+gamma: 0.99
+gae_lambda: 0.95
+num_minibatches: 4
+update_epochs: 4
+norm_adv: true
+clip_coef: 0.1
+clip_vloss: true
+ent_coef: 0.01
+vf_coef: 0.5
+max_grad_norm: 0.5
+target_kl: null

configs/ppo/dev_larger_timesteps_larger_rolloutsteps.yaml

@@ -0,0 +1,16 @@
+learning_rate: 0.0003
+total_timesteps: 1228800
+num_envs: 32
+num_steps: 64
+anneal_lr: true
+gamma: 0.99
+gae_lambda: 0.95
+num_minibatches: 32
+update_epochs: 4
+norm_adv: true
+clip_coef: 0.2
+clip_vloss: true
+ent_coef: 0.005
+vf_coef: 1.0
+max_grad_norm: 0.5
+target_kl: null

configs/ppo/smoke_test.yaml

@@ -0,0 +1,19 @@
+# Fast PPO Configuration
+# Lower timestep count for quick iterations/testing.
+
+learning_rate: 0.0005
+total_timesteps: 1024
+num_envs: 32
+num_steps: 32
+anneal_lr: true
+gamma: 0.99
+gae_lambda: 0.95
+num_minibatches: 4
+update_epochs: 4
+norm_adv: true
+clip_coef: 0.2
+clip_vloss: true
+ent_coef: 0.01
+vf_coef: 0.5
+max_grad_norm: 0.5
+target_kl: null

configs/ppo/stable.yaml

@@ -0,0 +1,19 @@
+# Stable PPO Configuration
+# Standard hyperparams with lower LR and larger batch.
+
+learning_rate: 0.0001
+total_timesteps: 10000000
+num_envs: 100
+num_steps: 256
+anneal_lr: true
+gamma: 0.99
+gae_lambda: 0.95
+num_minibatches: 8
+update_epochs: 4
+norm_adv: true
+clip_coef: 0.1
+clip_vloss: true
+ent_coef: 0.01
+vf_coef: 0.5
+max_grad_norm: 0.5
+target_kl: null

configs/ring-final.yaml

@@ -0,0 +1,74 @@
+# Custom Main Configuration
+#
+# Use with:
+#   uv run python scripts/train.py --config-name main_config_custom
+#
+# This keeps the project defaults intact while giving you a single custom
+# training entrypoint you can edit freely.
+
+defaults:
+  - brittle_star_config
+  - experiment: base
+  - logging: default
+  - evaluation: default
+  - ppo: default
+  - architecture: decentralized
+  - morphology: 5_arms_full
+  - arena: default
+  - environment: directed_locomotion
+  - simulation: default
+  - _self_
+
+architecture:
+  topology_type: "ring"
+
+morphology:
+  morph_mode: RING
+
+experiment:
+  exp_name: "final-models/ring/"
+  seed: 42
+  torch_deterministic: true
+  cuda: true
+
+logging:
+  track: true
+  save_model: true
+  save_checkpoints: true
+  upload_final_model: true
+  upload_checkpoints: true
+  checkpoint_frequency: 20
+  wandb_project_name: "final-models"
+
+evaluation:
+  evaluate_checkpoints: true
+  eval_max_steps: 2000
+  eval_seed: 0
+
+ppo:
+  learning_rate: 0.0001
+  total_timesteps: 16384000
+  num_envs: 128
+  num_steps: 64
+  anneal_lr: true
+  gamma: 0.99
+  gae_lambda: 0.95
+  num_minibatches: 32
+  update_epochs: 4
+  norm_adv: true
+  clip_coef: 0.2
+  clip_vloss: true
+  ent_coef: 0.001
+  vf_coef: 1.0
+  max_grad_norm: 0.5
+  target_kl: 0.02
+
+environment:
+  simulation_time: 100000.0
+  target_distance: 3.0
+
+hydra:
+  job:
+    chdir: true
+  run:
+    dir: ${experiment.base_run_dir}/${experiment.exp_name}/${now:%Y-%m-%d}/${now:%H-%M-%S}

configs/simulation/default.yaml

@@ -0,0 +1,26 @@
+# Default Simulation Settings
+# These values are used by scripts/simulate.py
+
+# Path to the trained model (optional)
+model_path: null
+
+# Script behavior
+headless: false
+# In headless mode this is required; in viewer mode null means "infinite".
+max_steps: null
+
+# Optional: override morphology for amputation experiments.
+# Points to a morphology config YAML file (e.g. configs/morphology/3_arms.yaml).
+# If null, the training morphology from the model's metadata is used.
+morphology_override: null
+
+# Video recording (requires [evaluation] extra)
+record_video: false
+# When null, video is saved in a per-model evaluation folder alongside the model.
+video_output_path: null
+# Camera ID to use for video recording (1 is usually the close-up camera)
+camera_id: 1
+
+# Optional override for the metadata YAML file path.
+# If null, the script looks for `<model_name>_metadata.yaml` alongside the model_path.
+metadata_path: null