[BUG] XVLA async inference error

[eliasab16]

I'm running the following async inference script:

server_address: localhost:8080

robot:
  type: so101_follower
  port: /dev/tty.usbmodem5A680122821
  id: right_follower
  cameras:
    wrist_top:
      type: opencv
      index_or_path: 0
      width: 800
      height: 600
      fps: 30
    overview_top:
      type: opencv
      index_or_path: 1
      width: 800
      height: 600
      fps: 30
    overview_bottom:
      type: opencv
      index_or_path: 2
      width: 800
      height: 600
      fps: 30
    wrist_bottom:
      type: opencv
      index_or_path: 3
      width: 800
      height: 600
      fps: 30


policy_type: xvla
policy_device: mps
actions_per_chunk: 50
chunk_size_threshold: 0.8
aggregate_fn_name: weighted_average
debug_visualize_queue_size: true
pretrained_name_or_path: {hf_username}/{name}
task: {task_name}

rename_map:
  observation.images.overview_bottom: observation.images.image
  observation.images.overview_top: observation.images.image2
  observation.images.wrist_bottom: observation.images.image3
  observation.images.wrist_top: observation.images.image4

But I'm running into the following error (not very descriptive):

INFO 2025-12-10 23:33:04 y_server.py:227 Running inference for observation #0 (must_go: False)
ERROR 2025-12-10 23:33:04 y_server.py:267 Error in StreamActions: 'observation.images.wrist_top'

I tried to replace the camera names in the "cameras" field to match the names used in the training, but I still run into the same error with just the camera name updated in the error message.

For context, the camera names are the ones I used when collecting the data, then I mapped them to the standardized format ("image", "image2", etc.) during training. I'm passing the exact format to the "rename_map" parameter*.

Resources consulted:
-> https://huggingface.co/docs/lerobot/en/introduction_processors.

*Note: I locally updated the async inference script to accept the "rename_map" parameter (the logic handling the mapping already exists, just the script isn't yet updated by lerobot to accept the param).