fixed env eval setup

ml/evaluate.py

@@ -13,7 +13,7 @@ def evaluate(
     model_path: str,
     episodes: int = 5,
     use_gui: bool = True,
-    num_robots: int = 1,
+    num_robots: int = 16,  # Default to 16 to match your trained (352,) observation space
     robot_spacing: float = 0.5,
     start_pose: str = "init_deg",
     random_command: bool = True,
@@ -29,9 +29,10 @@ def evaluate(
     from .env import JackBotEnv
 
     print(f"[Eval] Loading policy model from: {model_path}")
-    model = PPO.load(model_path)
+    # Force device="cpu" to prevent AMD ROCm/hipBLASLt matrix multiplication crashes
+    model = PPO.load(model_path, device="cpu")
 
-    # Initialize environment with active command sampling so robot actually walks
+    # Initialize standard environment (returns single array of shape (352,))
     env = JackBotEnv(
         use_gui=use_gui,
         random_command=random_command,
@@ -52,7 +53,6 @@ def evaluate(
         print(f"\n--- Starting Evaluation Episode {ep + 1}/{episodes} ---")
 
         while not done:
-            # Deterministic evaluation (no exploration noise)
             action, _ = model.predict(obs, deterministic=True)
             obs, reward, terminated, truncated, _ = env.step(action)
             
@@ -60,7 +60,6 @@ def evaluate(
             total_reward += float(reward)
             steps += 1
 
-            # Give PyBullet GUI frame pacing if running visually
             if use_gui:
                 time.sleep(1.0 / 240.0)
 
@@ -70,7 +69,6 @@ def evaluate(
 
     env.close()
 
-    # Calculate summary statistics
     metrics = {
         "model_path": str(model_path),
         "episodes_evaluated": episodes,
@@ -87,7 +85,6 @@ def evaluate(
     print(f"Mean Episode Length: {metrics['mean_episode_length']:.1f} steps")
     print("=" * 50)
 
-    # Optional JSON metrics export
     if save_json:
         out_path = Path(save_json)
         out_path.parent.mkdir(parents=True, exist_ok=True)
@@ -95,26 +92,4 @@ def evaluate(
             json.dump(metrics, f, indent=4)
         print(f"[Eval] Saved evaluation metrics to: {out_path.resolve()}")
 
-    return metrics
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description="Evaluate a trained JackBot policy.")
-    parser.add_argument("--model-path", type=str, required=True, help="Path to trained PPO model zip")
-    parser.add_argument("--episodes", type=int, default=5, help="Number of evaluation rounds")
-    parser.add_argument("--gui", action="store_true", help="Render GUI simulation")
-    parser.add_argument("--num-robots", type=int, default=1, help="Number of robots in sim")
-    parser.add_argument("--robot-spacing", type=float, default=0.5, help="Spacing between robots in meters")
-    parser.add_argument("--start-pose", type=str, choices=["init_deg", "init90_deg"], default="init_deg")
-    parser.add_argument("--save-json", type=str, default=None, help="Optional output JSON path for metrics")
-    args = parser.parse_args()
-
-    evaluate(
-        model_path=args.model_path,
-        episodes=args.episodes,
-        use_gui=args.gui,
-        num_robots=args.num_robots,
-        robot_spacing=args.robot_spacing,
-        start_pose=args.start_pose,
-        save_json=args.save_json,
-    )
+    return metrics