fix training

lerobot/common/envs/factory.py

@@ -1,5 +1,7 @@
 from torchrl.envs.transforms import StepCounter, TransformedEnv
 
+from rl.torchrl.envs.transforms.transforms import Compose, Transform
+
 
 def make_env(cfg, transform=None):
     kwargs = {
@@ -33,7 +35,13 @@ def make_env(cfg, transform=None):
 
     if transform is not None:
         # useful to add normalization
-        env.append_transform(transform)
+        if isinstance(transform, Compose):
+            for tf in transform:
+                env.append_transform(tf.clone())
+        elif isinstance(transform, Transform):
+            env.append_transform(transform.clone())
+        else:
+            raise NotImplementedError()
 
     return env