Minor + increase obs norm constants

baselines/ippo/ippo_pufferlib.py

@@ -339,5 +339,4 @@ def run(
 
 
 if __name__ == "__main__":
-
     app()

integrations/rl/puffer/puffer_env.py

@@ -340,7 +340,7 @@ def step(self, action):
         # (6) Get the next observations. Note that we do this after resetting
         # the worlds so that we always return a fresh observation
         next_obs = self.env.get_obs(self.controlled_agent_mask)
-        
+
         self.observations = next_obs
         self.rewards = reward_controlled
         self.terminals = terminal

pygpudrive/env/constants.py

@@ -7,8 +7,10 @@
 MAX_VEH_LEN = 30
 MAX_VEH_WIDTH = 15
 MAX_VEH_HEIGHT = 5
-MIN_REL_GOAL_COORD = -1000
-MAX_REL_GOAL_COORD = 1000
+MIN_REL_GOAL_COORD = (
+    -10_000
+)  # Note: (1000 or 100 should be enough, may indicate a bug or these are just from dead controlled agents)
+MAX_REL_GOAL_COORD = 10_000
 MIN_REL_AGENT_POS = -1000
 MAX_REL_AGENT_POS = 1000
 MAX_ORIENTATION_RAD = 2 * np.pi
@@ -21,5 +23,5 @@
 
 # Feature shape constants
 EGO_FEAT_DIM = 6
-PARTNER_FEAT_DIM = 6 
+PARTNER_FEAT_DIM = 6
 ROAD_GRAPH_FEAT_DIM = 13

pygpudrive/env/env_torch.py

@@ -373,8 +373,6 @@ def _get_ego_state(self, mask) -> torch.Tensor:
         if self.config.norm_obs:
             ego_state.normalize()
 
-        # TODO: I deleted this permute. Was it needed?
-        # .permute(1, 2, 0)
         return [ego_state.data]
         """
         [

pygpudrive/visualize/core.py

@@ -25,8 +25,6 @@
     AGENT_COLOR_BY_STATE,
 )
 
-import pdb
-
 OUT_OF_BOUNDS = 1000
 
 
@@ -209,12 +207,16 @@ def plot_simulator_state(
 
             if time_step is not None and not eval_mode:
                 # Plot rollout statistics
-                num_controlled = controlled.sum().item() 
+                num_controlled = controlled.sum().item()
                 num_off_road = is_offroad.sum().item()
                 num_collided = is_collided.sum().item()
-                off_road_rate = num_off_road / num_controlled if num_controlled > 0 else 0
-                collision_rate = num_collided / num_controlled if num_controlled > 0 else 0
-
+                off_road_rate = (
+                    num_off_road / num_controlled if num_controlled > 0 else 0
+                )
+                collision_rate = (
+                    num_collided / num_controlled if num_controlled > 0 else 0
+                )
+
                 ax.text(
                     0.5,  # Horizontal center
                     0.95,  # Vertical location near the top
@@ -229,19 +231,21 @@ def plot_simulator_state(
                     bbox=dict(facecolor="white", edgecolor="none", alpha=0.9),
                 )
             else:
-                if eval_mode and results_df is not None: 
-
-                    num_controlled = results_df.iloc[env_idx].controlled_agents_in_scene
+                if eval_mode and results_df is not None:
+
+                    num_controlled = results_df.iloc[
+                        env_idx
+                    ].controlled_agents_in_scene
                     off_road_rate = results_df.iloc[env_idx].off_road * 100
                     collision_rate = results_df.iloc[env_idx].collided * 100
                     goal_rate = results_df.iloc[env_idx].goal_achieved * 100
-                    other = results_df.iloc[env_idx].not_goal_nor_crashed * 100       
-                    
+                    other = results_df.iloc[env_idx].not_goal_nor_crashed * 100
+
                     ax.text(
                         0.5,  # Horizontal center
                         0.95,  # Vertical location near the top
-                        f"$N_c$ = {num_controlled}; " 
-                        f"OR: {off_road_rate:.1f}; " 
+                        f"t = {time_step} | $N_c$ = {num_controlled}; "
+                        f"OR: {off_road_rate:.1f}; "
                         f"CR: {collision_rate:.1f}; "
                         f"GR: {goal_rate:.1f}; "
                         f"Other: {other:.1f}",
@@ -250,7 +254,9 @@ def plot_simulator_state(
                         transform=ax.transAxes,
                         fontsize=20 * marker_scale,
                         color="black",
-                        bbox=dict(facecolor="white", edgecolor="none", alpha=0.9),
+                        bbox=dict(
+                            facecolor="white", edgecolor="none", alpha=0.9
+                        ),
                     )
 
             # Determine center point for zooming
@@ -268,7 +274,7 @@ def plot_simulator_state(
             # Set zoom window around the center
             ax.set_xlim(center_x - zoom_radius, center_x + zoom_radius)
             ax.set_ylim(center_y - zoom_radius, center_y + zoom_radius)
-            
+
             # Remove ticks
             ax.set_xticks([])
             ax.set_yticks([])