feat(PPOTrainer): decentralized message passing babyyyy

configs/architecture/decentralized.yaml

@@ -26,7 +26,7 @@ critic:
   activation: "tanh"
 
 # Synchronous message-passing rounds per control step
-message_passing_steps: 1
+message_passing_steps: 4
 
 # Connectivity topology (e.g., ring, fully_connected)
 topology_type: "ring"

configs/environment/dir_loc_further.yaml

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

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/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