turn off other penalties

kscalelabs · Apr 3, 2024 · e188cf7 · e188cf7
1 parent ec78264
commit e188cf7
Show file tree

Hide file tree

Showing 2 changed files with 11 additions and 5 deletions.
diff --git a/sim/humanoid_gym/envs/humanoid_config.py b/sim/humanoid_gym/envs/humanoid_config.py
@@ -159,10 +159,14 @@ class scales:  # noqa: N801
             base_acc = 0.0
 
             # Energy
-            torques = -1e-6
-            dof_vel = -5e-5
-            dof_acc = -1e-8
-            collision = -1e-2
+            # torques = -1e-6
+            # dof_vel = -5e-5
+            # dof_acc = -1e-8
+            # collision = -1e-2
+            torques = 0.0
+            dof_vel = 0.0
+            dof_acc = 0.0
+            collision = 0.0
 
     class normalization:  # noqa: N801
         class obs_scales:  # noqa: N801

diff --git a/sim/humanoid_gym/envs/humanoid_env.py b/sim/humanoid_gym/envs/humanoid_env.py
@@ -251,7 +251,9 @@ def _reward_base_height(self) -> Tensor:
         """
         base_height = self.root_states[:, 2]
         max_foot_height = self.rigid_state[:, self.feet_indices, 2].max(dim=1).values
-        return base_height - max_foot_height
+        height_diff = base_height - max_foot_height
+        height_diff[height_diff < 0.3] = 0.0  # No cheating.
+        return height_diff
 
     def _reward_base_acc(self) -> Tensor:
         """Computes the reward based on the base's acceleration.