Merge remote-tracking branch 'origin/develop-add-fire-debug' into develop-add-fire

ml-agents/mlagents/trainers/models_torch.py

         return normalized_state
 
     def update(self, vector_input):
-        mean_current_observation = vector_input.mean(0).type(torch.float32)
+        steps_increment = vector_input.size()[0]
+        total_new_steps = self.normalization_steps + steps_increment
+
+        input_to_old_mean = vector_input - self.running_mean
-            mean_current_observation - self.running_mean
-        ) / (self.normalization_steps + 1).type(torch.float32)
-        new_variance = self.running_variance + (mean_current_observation - new_mean) * (
-            mean_current_observation - self.running_mean
-        )
+            input_to_old_mean / total_new_steps.type(torch.float32)
+        ).sum(0)
+
+        input_to_new_mean = vector_input - new_mean
+        new_variance = self.running_variance + (
+            input_to_new_mean * input_to_old_mean
+        ).sum(0)
-        self.normalization_steps = self.normalization_steps + 1
+        self.normalization_steps = total_new_steps
 
 
 class ValueHeads(nn.Module):