Key Features

What you can do

Pretrained Models and Datasets

Provides pretrained models and human-collected demonstration datasets hosted on Hugging Face for immediate use without physical robot assembly.

Simulation Environments

Includes simulation environments that allow users to start robotics tasks without needing physical hardware.

Real Robot Data Recording and Control

Offers tools for recording data on physical robots and supports end-to-end control using Vision-Language-Action models.

Support for Imitation and Reinforcement Learning

Supports state-of-the-art imitation learning and reinforcement learning policies transferable to real-world robots.

Hardware Integration

Integrates with specific affordable hardware platforms like ROBOTIS OMX, Seeed Studio SO-ARM10x, and Trossen Robotics arms via forked repositories.

Standardized Dataset Format

Uses a standardized dataset format compatible with PyTorch and Hugging Face to facilitate data sharing and experiment reproducibility.