Converts an image to uint8 if it is a float image. This is important for reducing the size of the image when sending it over the network.

Return `quaternion` from Euler angles and axis sequence `axes` Parameters ---------- ai : float First rotation angle (according t

todo

TODO

Extract agentview and eye_in_hand images from raw env obs dict. Avoids torch round-trip: the env already returns uint8 numpy arrays [H, W, C

Add a black title bar with text above the image

Convert quaternion to Euler angles (roll, pitch, yaw) quat: [rx, ry, rz, rw] format Return: [roll, pitch, yaw] (radians)

Main training loop - train by steps instead of epochs.

Write data to a JSON file. Creates parent directories if they don't exist. Args: data (dict): The dictionary to write. fpath

Get next batch from iterator, reset if epoch is finished.

Prepare input dict following infer code pattern from wan_va_server.py.

Replicates tf.image.resize_with_pad for one image using PIL. Resizes an image to a target height and width without distortion by padding with zero

Train a single batch, returns losses for logging.

Apply activation checkpointing to the model.

Convert input dict to match transformer input format if needed.

TODO

r""" TODO

Generates an action with the VLA policy.

TODO: show be rewritten as a dict

Parse arguments and run training.

TODO