diff --git a/flatland/envs/predictions.py b/flatland/envs/predictions.py
index 0420ab70737503ec135eb05881b4b9a82688f9df..ecc0f8f18d3a2d9cb730c9fcadf237e7ce133a6d 100644
--- a/flatland/envs/predictions.py
+++ b/flatland/envs/predictions.py
@@ -4,6 +4,7 @@ Collection of environment-specific PredictionBuilder.
 
 import numpy as np
 
+from envs.rail_env import RailEnvActions
 from flatland.core.env_prediction_builder import PredictionBuilder
 
 
@@ -41,12 +42,7 @@ class DummyPredictorForRailEnv(PredictionBuilder):
         prediction_dict = {}
 
         for agent in agents:
-
-            # 0: do nothing
-            # 1: turn left and move to the next cell
-            # 2: move to the next cell in front of the agent
-            # 3: turn right and move to the next cell
-            action_priorities = [2, 1, 3]
+            action_priorities = [RailEnvActions.MOVE_FORWARD, RailEnvActions.MOVE_LEFT, RailEnvActions.MOVE_RIGHT]
             _agent_initial_position = agent.position
             _agent_initial_direction = agent.direction
             prediction = np.zeros(shape=(self.max_depth, 5))