Robots Atlas>ROBOTS ATLAS
Helix
Figure AI's generalist Vision-Language-Action (VLA) model unifying perception, language understanding and real-time humanoid robot control.
✓ Active🔬 Research onlyLLMModel sterowania robotemModel multimodalny
Parameters
System 2: 7B + System 1: 80M
parameters
Release date
20 February 2025
Access:HostedDeployment:📱 On-device

Overview

Helix is a generalist Vision-Language-Action (VLA) model developed by Figure AI that unifies perception, language understanding and learned control of a humanoid robot. It is the first VLA to output continuous control of the entire humanoid upper body at 200 Hz — including wrists, torso, head and individual fingers. It enables zero-shot generalization (grasping thousands of novel objects without prior training on them), multi-robot collaboration using identical network weights and natural-language conditioning. Its architecture uses a "System 1, System 2" design: System 2 (7B parameters) is an internet-pretrained VLM running at 7-9 Hz for scene understanding, and System 1 (80M parameters) is a transformer-based visuomotor policy running at 200 Hz. The model runs entirely on embedded low-power GPUs and was trained on approximately 500 hours of high-quality teleoperated data.

Classification
LLMLLMLLM
Access & deployment
Hosted
On-device
Weights: Closed
Key parameters
🧩 Parameters: System 2: 7B + System 1: 80M
📥 Input: image, text, robot sensors, robot state data

Technical specification

Parameters
System 2: 7B + System 1: 80M
parameters
Modalities
⬇ Input
imagetextrobot_sensorsrobot_state_data
⬆ Output
robot_actionsmanipulator_controlmotion_trajectories