Fix: Adaptive gripper threshold with debug logging

app.py

@@ -129,7 +129,43 @@ def predict(inst, img1, img2):
                     'qz': float(ac[6])
                 })
             
-            grip = [int(g > 0.5) for g in gripper_np]
+            # Parse gripper with adaptive normalization
+            gripper_np_flat = gripper_np.flatten()
+            
+            # Debug: Print gripper statistics
+            print(f"   📊 Gripper stats:")
+            print(f"      min={gripper_np_flat.min():.4f}, max={gripper_np_flat.max():.4f}")
+            print(f"      mean={gripper_np_flat.mean():.4f}, std={gripper_np_flat.std():.4f}")
+            print(f"      sample: {gripper_np_flat[:5]}")
+            
+            # Normalize gripper values if they have variation
+            g_range = gripper_np_flat.max() - gripper_np_flat.min()
+            
+            if g_range > 0.1:
+                # Significant variation: Normalize to [0, 1] range
+                gripper_norm = (gripper_np_flat - gripper_np_flat.min()) / g_range
+                grip = [int(g > 0.5) for g in gripper_norm]
+                print(f"   ✓ Normalized (range={g_range:.4f})")
+            elif g_range > 0.01:
+                # Small variation: Use adaptive median threshold
+                threshold = np.median(gripper_np_flat)
+                grip = [int(g > threshold) for g in gripper_np_flat]
+                print(f"   ✓ Adaptive threshold={threshold:.4f}")
+            else:
+                # Almost no variation
+                mean_val = gripper_np_flat.mean()
+                if mean_val > 0.7:
+                    grip = [int(g > 0.9) for g in gripper_np_flat]
+                    print(f"   ✓ High values (mean={mean_val:.4f}), threshold=0.9")
+                elif mean_val < 0.3:
+                    grip = [int(g > 0.1) for g in gripper_np_flat]
+                    print(f"   ✓ Low values (mean={mean_val:.4f}), threshold=0.1")
+                else:
+                    grip = [0] * len(gripper_np_flat)
+                    print(f"   ⚠ No variation, defaulting to OPEN")
+            
+            print(f"   🤖 Result: {sum(grip)}/{len(grip)} CLOSED")
+
             
             traj_plot = plot_traj(acts)
             grip_plot = plot_grip(grip)