Merge branch 'dev' into feat/wandb-logging

configs/production_training.yaml

@@ -4,34 +4,33 @@
 # with wandb logging enabled.
 
 # Experiment settings
-exp_name: "brittle_star_production"
+exp_name: "brittle_star_production_training"
 seed: 42
 
-# Tracking settings - IMPORTANT: Set your own wandb_entity!
+# Tracking
 track: true
+capture_video: false
 wandb_project_name: "PPO-Modularity"
-wandb_entity: "SEL3-2026-Groep-4"  # ⚠️  SET THIS TO YOUR WANDB USERNAME OR TEAM
+wandb_entity: "SEL3-2026-Groep-4"
 
 # Model saving
 save_model: true
 checkpoint_frequency: 100  # Save checkpoint every 100 iterations
 
 # Environment settings
-num_envs: 32  # Increased for production
+num_envs: 512
 
-# Training hyperparameters - Production scale
-total_timesteps: 50000000  # 50M timesteps for full training
-learning_rate: 0.00025
-num_steps: 256  # Longer rollouts
+# Training hyperparameters
+total_timesteps: 50000000
+num_steps: 256
+num_minibatches: 4
+update_epochs: 4
+
+learning_rate: 2.5e-4
 anneal_lr: true
-
-# PPO specific - Fine-tuned
 gamma: 0.99
 gae_lambda: 0.95
-num_minibatches: 8  # More minibatches for stability
-update_epochs: 4
-norm_adv: true
-clip_coef: 0.2
+clip_coef: 0.1
 clip_vloss: true
 ent_coef: 0.01
 vf_coef: 0.5