# FAST-ACT

**Repository Path**: boxjod/FAST-ACT

## Basic Information

- **Project Name**: FAST-ACT
- **Description**: A super-fast version of the Action Chunking Transformer, paper in progress, coming soon!
- **Primary Language**: Unknown
- **License**: BSD-3-Clause
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2024-10-13
- **Last Updated**: 2024-10-13

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

### Declaration

With a robot action generation model based on a neurological conceptualization technique of **action chunking**, but existing robotic demonstration videos and academic work are executed at **4X or 8X speedups** before reaching normal human levels. Our work has the advantage of **state-of-the-art execution speeds** that can dramatically increase the speed of task execution while maintaining the smoothness and success of the baseline model trajectories.

We thank the [aloha act](https://github.com/tonyzhaozh/act) series for the foundation of the work, as well as the ARX Robotics Research Center and Faculty of Robotics Science and Engineering of Northeastern University for their sponsorship and technical support. This work was done during an internship at ARX.