diff --git a/tests/test_multi_speed.py b/tests/test_multi_speed.py
index 5918c24eb413e57a6b5d9ddb4cd3ff1f461e5c29..561057d81b431dfbb87b904f7a57e6fcbf84f84e 100644
--- a/tests/test_multi_speed.py
+++ b/tests/test_multi_speed.py
@@ -8,6 +8,7 @@ from flatland.envs.rail_generators import sparse_rail_generator, rail_from_grid_
 from flatland.envs.line_generators import sparse_line_generator
 from flatland.utils.simple_rail import make_simple_rail
 from test_utils import ReplayConfig, Replay, run_replay_config, set_penalties_for_replay
+from flatland.envs.agent_utils import RailAgentStatus
 
 
 # Use the sparse_rail_generator to generate feasible network configurations with corresponding tasks
@@ -49,7 +50,7 @@ class RandomAgent:
 def test_multi_speed_init():
     env = RailEnv(width=50, height=50,
                   rail_generator=sparse_rail_generator(seed=2), line_generator=sparse_line_generator(),
-                  random_seed=2,
+                  random_seed=3,
                   number_of_agents=3)
     
     # Initialize the agent with the parameters corresponding to the environment and observation_builder
@@ -62,13 +63,17 @@ def test_multi_speed_init():
     # Reset environment and get initial observations for all agents
     env.reset(False, False)
 
+    for a_idx in range(len(env.agents)):
+        env.agents[a_idx].position =  env.agents[a_idx].initial_position
+        env.agents[a_idx].status = RailAgentStatus.ACTIVE
+
     # Here you can also further enhance the provided observation by means of normalization
     # See training navigation example in the baseline repository
     old_pos = []
     for i_agent in range(env.get_num_agents()):
-        env.agents[i_agent].speed_data['speed'] = 1. / (i_agent + 2)
+        env.agents[i_agent].speed_data['speed'] = 1. / (i_agent + 1)
         old_pos.append(env.agents[i_agent].position)
-
+        print(env.agents[i_agent].position)
     # Run episode
     for step in range(100):