diff --git a/setup.py b/setup.py
index fa1d03f31..5a1846697 100644
--- a/setup.py
+++ b/setup.py
@@ -203,11 +203,11 @@ def get_local_version(version: "ScmVersion", time_format="%Y%m%d") -> str:
         "torch>=1.4.0",
         "tqdm",
         "scikit-learn>=0.21.2",
-        "seals>=0.1.5",
+        "seals~=0.1.5",
         STABLE_BASELINES3,
         "sacred>=0.8.4",
         "tensorboard>=1.14",
-        "huggingface_sb3>=2.2.1",
+        "huggingface_sb3~=2.3",
         "datasets>=2.8.0",
     ],
     tests_require=TESTS_REQUIRE,
diff --git a/tests/algorithms/test_mce_irl.py b/tests/algorithms/test_mce_irl.py
index 53b9527df..9fd527c41 100644
--- a/tests/algorithms/test_mce_irl.py
+++ b/tests/algorithms/test_mce_irl.py
@@ -132,6 +132,8 @@ def test_infinite_horizon_error(random_mdp, rng):
 def test_policy_om_random_mdp(discount: float):
     """Test that optimal policy occupancy measure ("om") for a random MDP is sane."""
     mdp = gym.make("seals/Random-v0")
+    mdp.seed(0)
+
     V, Q, pi = mce_partition_fh(mdp, discount=discount)
     assert np.all(np.isfinite(V))
     assert np.all(np.isfinite(Q))