• I have marked all applicable categories:
  • exception-raising bug
  • RL algorithm bug
  • documentation request (i.e. "X is missing from the documentation.")
  • new feature request
• I have visited the source website, and in particular read the known issues
• I have searched through the issue categories for duplicates
• I have mentioned version numbers, operating system and environment, where applicable:

  import tianshou, torch, sys
  print(tianshou.__version__, torch.__version__, sys.version, sys.platform)

Related to issue #27 : I have a multimodal observation space that consists of image, optical flow, segmentation mask (int array), and other low-dimensional states. The current obs seems to support only a single uniform tensor. Is there any way to have tuple/dict obs that gets collated along the batch dimension? For example, pytorch provides collate_fn that preserves data structure: https://pytorch.org/docs/stable/data.html#dataloader-collate-fn
For now I can put the other obs in info, but I have to manually batch them at run time, which can be tedious and error-prone.