Act Dataset Training is an imitation learning model that uses a method called Action Chunking with Transformers (ACT) to predict short sequences of actions, rather than individual steps. It is designed to learn from teleoperated data, with the goal of producing policies that can achieve high success rates in relevant tasks. The model is associated with robotics applications and has been trained and made available on the Hugging Face platform.
This model can be trained from scratch and supports evaluation or inference on new data. Instructions are provided for using Act Dataset Training with various libraries, including LeRobot, as well as with platforms such as Google Colab and Kaggle. The model is distributed in the "safetensors" format and is accompanied by documentation and guides for training and running inference.
Act Dataset Training is released under the Apache-2.0 license. It is accessible for use through the Hugging Face Hub, and its documentation references additional resources and guides for users interested in robotics and imitation learning research.
Act Dataset Training is an Other AI product. It focuses on automating the prediction of action sequences from teleoperated data for robotics and imitation learning tasks. It is built as an open-source project for machine learning researchers. Act Dataset Training is open source under the Open Source license. The product ships for the web and the command line.
It is developed by aimandanial, and the product first shipped in 2023. Key capabilities include imitation learning, action chunk prediction, and transformer model.
Latest indexed changes and source events
aimandanial/act_dataset_training verified by the PulseGate indexer
Other apps tracked under the same category.
