Figure AI has announced Helix, a Vision-Language-Action (VLM) model that allows humanoid robots to perform complex tasks using natural language.
This model marks progress in robotics, enabling robots to understand and react to instructions in real time, handle unforeseen objects, and collaborate.
“We’ve been working on this project for over a year, aiming to solve general robotics. Like a human, Helix understands speech, reasons through problems, and can grasp any object – all without needing training or code,” said Brett Adcock, founder at Figure AI.
Meet Helix, our in-house AI that reasons like a human
— Figure (@Figure_robot) February 20, 2025
Robotics won't get to the home without a step change in capabilities
Our robots can now handle virtually any household item: pic.twitter.com/Wsx5s8Qelc
This development comes after Adcock announced “the decision to leave our Collaboration Agreement with OpenAI” in his X post on February 5th. OpenAI has been an investor in the company before.
In the same post, he had also hinted at this Helix development saying, “We’re excited to show you in the next 30 days something no one has ever seen on a humanoid.”
In another recent post, Adcock also said this is the breakthrough year for AI Robotics we’ve been waiting for. With Helix, he mentioned that 2025 will be a pivotal year as the company starts production, ships more robots, and tackles home robotics.
What is Helix?
In the video posted by the company, one Helix neural network is running on 2 Figure robots at once. The company said, “Our robots equipped with Helix can now pick up virtually any household object without any code or prior training.”
Both the robots can achieve coordination through natural language prompts like “Hand the bag of cookies to the robot on your right” or “Receive the bag of cookies from the robot on your left and place it in the open drawer.”
Helix is a “System 1, System 2” VLA model for high-rate, dexterous control of the entire humanoid upper body.
It is the first VLA model to control a humanoid upper body, facilitate multi-robot collaboration, and pick up any small household object.
The model uses one set of neural network weights to learn behaviours without task-specific fine-tuning. It runs on low-power GPUs, making it commercially viable.
According to Figure AI in their technical paper, the current state of robotics requires “either hours of PhD-level expert manual programming or thousands of demonstrations” to teach robots new behaviours.
Looking Forward to Scaling Helix
“Helix matches the speed of specialised single-task behavioural cloning policies while generalising zero-shot to thousands of novel test objects,” said the company.
The company also notes that Helix achieves “strong performance across diverse tasks with a single unified model.”
The model was trained using approximately 500 hours of teleoperated behaviours. During training, the VLM processes video clips from the onboard robot cameras, prompted with: “What instruction would you have given the robot to get the action seen in this video?”
Figure AI is also looking forward to scaling Helix and inviting people to join their team.
As for OpenAI, according to its career page, the startup is hiring for roles in mechanical engineering, robotics systems integration, and program management. The goal is to “integrate cutting-edge hardware and software to explore a broad range of robotic form factors”.
Last year, the company hired Caitlin Kalinowski to lead its robotics and consumer hardware divisions. Previously at Meta, she oversaw the development of Orion augmented reality (AR) glasses. OpenAI has also invested in robotics AI startup Physical Intelligence.
Recently, tech giants Meta and Apple have also expressed plans to scale into developing humanoid robots.
