OPEN MODELS/IMITATION LEARNING/LEROBOT ACT
BEGINNERIMITATION LEARNING

LeRobot ACT

Action Chunking Transformer for bimanual manipulation

WHAT DOES THIS ACTUALLY DO?

Most robot control models make one decision at a time. ACT is smarter — it plans ahead and outputs a chunk of future moves at once. This makes it much smoother at two-handed tasks like folding clothes or doing dishes.

PARAMETERS
~50M
RELEASED
2024-03
LICENSE
Apache 2.0
ARCHITECTURE
Action...
CAN I ACTUALLY RUN THIS?
9
Beginner Friendly
OUT OF 10

One of the easiest entry points into robot learning. The LeRobot library makes training and inference straightforward. If you have a cheap robot arm, you can be up and running in an afternoon.

DOWNLOAD WEIGHTS
Download from Hugging FaceView on GitHub
1GB disk space required
OVERVIEW

Action Chunking with Transformers — a policy that predicts sequences of actions at once rather than one at a time, trained on ALOHA bimanual manipulation data.

TRAINED ON

ALOHA bimanual manipulation demonstrations (100+ tasks)

ARCHITECTURE
Action Chunking Transformer (ACT) + CVAE
CAPABILITIES
Bimanual coordination
Temporal action chunking
High-frequency control
Simulation and real-robot training
REAL ROBOTICS USE CASES
HOME ROBOTICS

Fold a shirt with two robot arms

EDUCATION

Train a LeRobot SO-101 arm in minutes

RESEARCH

Baseline for bimanual manipulation benchmarks

HARDWARE REQUIREMENTS
MIN VRAM
4GB
REC VRAM
8GB
MIN RAM
8GB
DISK SPACE
1GB
JETSON
Compatible
CPU ONLY
Supported
TESTED GPUs
RTX 3060RTX 4060M1/M2 MacAny 4GB GPU
DEPLOYMENT PATHS
Local PC
Runs on 4GB VRAMSUPPORTED
Docker
SUPPORTED
ROS2
SUPPORTED
Jetson Orin Nano
SUPPORTED
Raspberry Pi
UnderpoweredNOT SUPPORTED
Cloud
SUPPORTED
RELATED PAPERS
Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware
RELATED ORGANIZATIONS
Hugging FaceTrossen Robotics
ORGANIZATIONS
Hugging FaceStanford (ALOHA)
DIFFICULTY
BEGINNER

Anyone with a computer can run this

COMPUTE8/10
SETUP8/10
BEGINNER-FRIENDLY9/10
EDGE DEPLOYMENT
Jetson Compatible
CPU-only Supported
Community →

Seeded and planned prompts — not live forum activity yet.