Hi,
I'm trying to use the RT-1 Tensorflow code base for training. Following the Colab notebook for data loading I ended up with the following code
def to_trajectory(step):
# map open-x trajectory to tf_agent Trajectory
return trajectory.from_episode(
observation=step["observation"],
action=step["action"],
policy_info=(),
reward=tf.zeros((SEQUENCE_LENGTH,)),
discount=tf.ones((SEQUENCE_LENGTH,)),
)
# same as in the Colab open-x data loading
trajectory_dataset = trajectory_transform.transform_episodic_rlds_dataset(episodic_dataset)
trajectory_iter = iter(trajectory_dataset)
t = to_trajectory(next(trajectory_iter))
But when I use the t to train the RT-1 sequence agent I get and tensor spec mismatch
Hi, I'm trying to use the RT-1 Tensorflow code base for training. Following the Colab notebook for data loading I ended up with the following code
But when I use the
t
to train the RT-1 sequence agent I get and tensor spec mismatchAny ideas ?