diff --git a/grid2op/Environment.py b/grid2op/Environment.py
index 903605bea..7a10ac333 100644
--- a/grid2op/Environment.py
+++ b/grid2op/Environment.py
@@ -206,7 +206,8 @@ def __init__(self,
                  observationClass=CompleteObservation,
                  rewardClass=FlatReward,
                  legalActClass=AllwaysLegal,
-                 epsilon_poly=1e-2):
+                 epsilon_poly=1e-2,
+                 tol_poly=1e-6):
         """
         Initialize the environment. See the descirption of :class:`grid2op.Environment.Environment` for more information.
 
@@ -233,6 +234,7 @@ def __init__(self,
         self._time_powerflow = 0
         self._time_extract_obs = 0
         self._epsilon_poly = epsilon_poly
+        self._tol_poly = tol_poly
 
         # define logger
         self.logger = None
@@ -652,7 +654,7 @@ def _aux_redisp(self, redisp_act, target_p, avail_gen, previous_redisp):
             except_ = InvalidRedispatching("Impossible to perform this redispatching. Maximum ramp (or pmax) for "
                                            "available generators is not enough to absord "
                                            "{}MW, max possible is {}MW".format(val_sum, max_disp))
-        elif np.abs(val_sum) <= 1e-6:
+        elif np.abs(val_sum) <= self._tol_poly:
             # i don't need to modify anything so i should be good
             new_redisp = 0.0 * redisp_act
         else:
@@ -847,8 +849,8 @@ def _make_redisp_0sum(self, action, new_p):
 
         # get the target redispatching (cumulation starting from the first element of the scenario)
         self.target_dispatch += redisp_act_orig
-        if np.abs(np.sum(self.actual_dispatch)) >= 1e-6 or \
-                np.sum(np.abs(self.actual_dispatch - self.target_dispatch)) >= 1e-6:
+        if np.abs(np.sum(self.actual_dispatch)) >= self._tol_poly or \
+                np.sum(np.abs(self.actual_dispatch - self.target_dispatch)) >= self._tol_poly:
             # make sure the redispatching action is zero sum
             new_redisp, except_ = self._get_redisp_zero_sum(self.target_dispatch,
                                                             self.gen_activeprod_t,
@@ -1280,6 +1282,7 @@ def get_kwargs(self):
         res["rewardClass"] = self.rewardClass
         res["legalActClass"] = self.legalActClass
         res["epsilon_poly"] = self._epsilon_poly
+        res["tol_poly"] = self._tol_poly
         return res
 
     def get_params_for_runner(self):
diff --git a/grid2op/data/test_PandaPower/prods_charac.csv b/grid2op/data/test_PandaPower/prods_charac.csv
index e34e749d5..099c89548 100644
--- a/grid2op/data/test_PandaPower/prods_charac.csv
+++ b/grid2op/data/test_PandaPower/prods_charac.csv
@@ -1,6 +1,6 @@
 Pmax,Pmin,name,type,bus,max_ramp_up,max_ramp_down,min_up_time,min_down_time,marginal_cost,shut_down_cost,start_cost,x,y,V
-170.,0.,gen_0_4,thermal,170.,170.,170,0,0,10.0,0.,0.0,0,199,106.0
-60.,0.,gen_2_1,nuclear,3,60.,60,0.0,1.0,0.0,0,0.0,180,10,104.5
-100.,0.,gen_5_2,nuclear,3,100.,100.,0.0,0.0,0.0,0,0.0,646,10,101.0
-100.,0.,gen_7_3,nuclear,3,100.,100.,0.0,0.0,0.0,0,0.0,216,334,107.0
-40.,0.,gen_1_0,solar,3,40.,40.,0.0,0.0,0.0,0,0.0,718,280,109.0
+190.,0.,gen_0_4,thermal,170.,190.,190.,0,0,10.0,0.,0.0,0,199,106.0
+80.,0.,gen_2_1,nuclear,3,80.,80.,0.0,1.0,0.0,0,0.0,180,10,104.5
+120.,0.,gen_5_2,nuclear,3,120.,120.,0.0,0.0,0.0,0,0.0,646,10,101.0
+120.,0.,gen_7_3,nuclear,3,120.,120.,0.0,0.0,0.0,0,0.0,216,334,107.0
+50.,0.,gen_1_0,solar,3,50.,50.,0.0,0.0,0.0,0,0.0,718,280,109.0
diff --git a/grid2op/tests/test_Environment.py b/grid2op/tests/test_Environment.py
index 33fa1e1a7..91e14abb5 100644
--- a/grid2op/tests/test_Environment.py
+++ b/grid2op/tests/test_Environment.py
@@ -172,7 +172,7 @@ def test_reward(self):
             cp.disable()
             cp.print_stats(sort="tottime")
         assert i == 287, "Wrong number of timesteps"
-        assert np.abs(cum_reward - 5739.929117641016) <= self.tol_one, "Wrong reward"
+        assert np.abs(cum_reward - 5739.92911) <= self.tol_one, "Wrong reward"
 
 
 class TestIllegalAmbiguous(unittest.TestCase):
@@ -184,7 +184,7 @@ class TestIllegalAmbiguous(unittest.TestCase):
     def setUp(self):
         # powergrid
         self.tolvect = 1e-2
-        self.tol_one = 1e-5
+        self.tol_one = 1e-4
         self.env = make("case5_example")
 
     def compare_vect(self, pred, true):
diff --git a/grid2op/tests/test_RedispatchEnv.py b/grid2op/tests/test_RedispatchEnv.py
index 68db4f4ba..ffc4ef155 100644
--- a/grid2op/tests/test_RedispatchEnv.py
+++ b/grid2op/tests/test_RedispatchEnv.py
@@ -110,7 +110,7 @@ def test_basic_redispatch_act(self):
         act = self.env.action_space({"redispatch": [2, 5]})
         obs, reward, done, info = self.env.step(act)
         assert np.abs(np.sum(self.env.actual_dispatch)) <= self.tol_one
-        th_dispatch = np.array([0., -1.26379812,  5.,  0., -3.73620188])
+        th_dispatch = np.array([0., -1.44301856,  5.,  0., -3.55698144])
         assert self.compare_vect(self.env.actual_dispatch, th_dispatch)
         target_val = self.chronics_handler.real_data.prod_p[1, :] + self.env.actual_dispatch
         assert self.compare_vect(obs.prod_p[:-1], target_val[:-1])  # I remove last component which is the slack bus
@@ -124,10 +124,10 @@ def test_basic_redispatch_act(self):
     def test_redispatch_act_above_pmax(self):
         # in this test, the asked redispatching for generator 2 would make it above pmax, so the environment
         # need to "cut" it automatically, without invalidating the action
-        act = self.env.action_space({"redispatch": [2, 20]})
+        act = self.env.action_space({"redispatch": [2, 60]})
         obs, reward, done, info = self.env.step(act)
         assert np.abs(np.sum(self.env.actual_dispatch)) <= self.tol_one
-        th_dispatch = np.array([0.,  -4.64121498,  20., 0., -15.35878502])
+        th_dispatch = np.array([0., -10.46650072,  50.89066718,   0., -40.42416646])
         assert self.compare_vect(self.env.actual_dispatch, th_dispatch)
         target_val = self.chronics_handler.real_data.prod_p[1, :] + self.env.actual_dispatch
         assert self.compare_vect(obs.prod_p[:-1], target_val[:-1])  # I remove last component which is the slack bus
@@ -219,7 +219,7 @@ def test_redispacth_twice_same(self):
         obs, reward, done, info = self.env.step(act)
         assert np.all(obs.target_dispatch == np.array([ 0.,  0., 5.,  0.,  0.]))
         assert np.abs(np.sum(obs.actual_dispatch)) <= self.tol_one
-        assert self.compare_vect(obs.actual_dispatch, np.array([0., -1.26379812,  5.,  0., -3.73620188]))
+        assert self.compare_vect(obs.actual_dispatch, np.array([ 0., -1.44301856,  5.,  0., -3.55698144]))
         assert np.all(obs.prod_p <= self.env.gen_pmax)
         assert np.all(obs.prod_p >= self.env.gen_pmin)
 
@@ -227,7 +227,7 @@ def test_redispacth_twice_same(self):
         obs, reward, done, info = self.env.step(act)
         assert np.all(obs.target_dispatch == np.array([ 0.,  0., 10.,  0.,  0.]))
         assert np.abs(np.sum(obs.actual_dispatch)) <= self.tol_one
-        assert self.compare_vect(obs.actual_dispatch, np.array([0., -2.45340249, 10.,  0., -7.54659751]))
+        assert self.compare_vect(obs.actual_dispatch, np.array([0., -2.81339987, 10.,  0., -7.18660013]))
         assert np.all(obs.prod_p <= self.env.gen_pmax)
         assert np.all(obs.prod_p >= self.env.gen_pmin)
 
@@ -236,19 +236,19 @@ def test_redispacth_secondabovepmax(self):
         obs, reward, done, info = self.env.step(act)
         assert np.all(obs.target_dispatch == np.array([0.,  0., 20.,  0.,  0.]))
         assert np.abs(np.sum(obs.actual_dispatch)) <= self.tol_one
-        assert self.compare_vect(obs.actual_dispatch, np.array([0.,  -4.64121498,  20.,   0., -15.35878502]))
+        assert self.compare_vect(obs.actual_dispatch, np.array([0.,  -5.34776078,  20.,   0., -14.65223922]))
         assert np.all(obs.prod_p <= self.env.gen_pmax)
         assert np.all(obs.prod_p >= self.env.gen_pmin)
 
-        act = self.env.action_space({"redispatch": [(2, 20.)]})
+        act = self.env.action_space({"redispatch": [(2, 40.)]})
         obs, reward, done, info = self.env.step(act)
-        assert np.all(obs.target_dispatch == np.array([0.,  0., 40.,  0.,  0.]))
-        assert self.compare_vect(obs.actual_dispatch, np.array([0., -5.77603921, 30.39084553, 0., -24.61480632]))
+        assert np.all(obs.target_dispatch == np.array([0.,  0., 60.,  0.,  0.]))
+        assert self.compare_vect(obs.actual_dispatch, np.array([0., -10.31164036, 50.39070301, 0., -40.07906265]))
         assert np.all(obs.prod_p[:-1] <= self.env.gen_pmax[:-1])
         assert np.all(obs.prod_p[:-1] >= self.env.gen_pmin[:-1])
 
 
-# TODO test that if i try to redispatched a turned off generator it breaks everything
+# # TODO test that if i try to redispatched a turned off generator it breaks everything
 class TestLoadingBackendPandaPower(unittest.TestCase):
     def setUp(self):
         # powergrid