TensorFlow

Google's open-source deep learning framework; robotics datasets often use its data format.

TensorFlow is the deep learning framework Google Brain open-sourced in 2015, with version 2.x adopting Keras as its high-level interface. Academic model training has largely shifted to PyTorch and JAX since then, but TensorFlow still shows up regularly in robotics: Google's RT-1 and other work is built on it, and datasets such as Open X-Embodiment are stored in the RLDS format as TFRecord files, read through TensorFlow Datasets (TFDS); the data-loading pipelines for Octo and OpenVLA also depend on tf.data. So even when training a VLA in PyTorch, TensorFlow usually still needs to be installed just to read the data.

ExampleUse tfds.builder to load the BridgeData V2 subset from Open X-Embodiment, then convert it to PyTorch tensors to feed into fine-tuning OpenVLA.

Also called

TF

Related

PyTorch · JAX · TensorFlow Datasets (TFDS) · RLDS (Reinforcement Learning Datasets) · TFRecord · Open X-Embodiment

Sources

TensorFlow 官网 (Chinese)
TensorFlow Datasets