Make sure policies don't mutate the batch (#323)

lerobot/common/policies/vqbet/modeling_vqbet.py

@@ -98,6 +98,7 @@ class VQBeTPolicy(nn.Module, PyTorchModelHubMixin):
         """
 
         batch = self.normalize_inputs(batch)
+        batch = dict(batch)  # shallow copy so that adding a key doesn't modify the original
         batch["observation.images"] = torch.stack([batch[k] for k in self.expected_image_keys], dim=-4)
         # Note: It's important that this happens after stacking the images into a single key.
         self._queues = populate_queues(self._queues, batch)
@@ -123,6 +124,7 @@ class VQBeTPolicy(nn.Module, PyTorchModelHubMixin):
     def forward(self, batch: dict[str, Tensor]) -> dict[str, Tensor]:
         """Run the batch through the model and compute the loss for training or validation."""
         batch = self.normalize_inputs(batch)
+        batch = dict(batch)  # shallow copy so that adding a key doesn't modify the original
         batch["observation.images"] = torch.stack([batch[k] for k in self.expected_image_keys], dim=-4)
         batch = self.normalize_targets(batch)
         # VQ-BeT discretizes action using VQ-VAE before training BeT (please refer to section 3.2 in the VQ-BeT paper https://arxiv.org/pdf/2403.03181)