Figure AI’s humanoid robots are now performing logistics tasks using the startup’s self-developed Helix AI system.
The Silicon Valley-based startup just shared footage showing several Figure 02 robots performing sorting tasks at an unnamed warehouse. Founder and CEO Brett Adcock said on social media that it took only 30 days to deploy the humanoids on-site for Figure’s second customer.
The customer has not been named, but observers have speculated it’s Amazon or Wal-Mart. Adcock has said that, with the new customer, he sees a path toward manufacturing 100,000 Figure humanoid robots over the next four years.
The demo shows the robots sorting packages through their onboard camera feeds. The robots identify and position barcodes and self-correct mistakes. Adcock said the robots are using Helix’s Sport Mode, which accelerates their speed by 50 percent.
According to Figure, the robots have 3D depth perception, with head cameras that capture both fine details and full-scene context for better control. With visual proprioception, the robots can inspect their own arms and hands for self-calibration, which Figure says minimizes downtime and ensures consistent performance across multiple units.
The Helix system’s cross-robot policy transfer means Figure humanoids can share what they learn without needing individual calibration. Each robot can adjust its own sensor readings and joint movements in real-time. Once a control policy is trained on one robot, it can be applied across multiple robots without extensive fine-tuning. This drastically reduces the time and effort required to deploy a fleet of robots.
Figure announced its in-house AI system last week after severing ties with OpenAI. The company is reportedly in advanced talks to raise $1.5 billion in series C funding for a valuation of nearly $40 billion. If accurate, the amount is a 1.4K percent increase from its 2024 valuation of $2.6 billion.
Figure calls Helix a vision language model (VLA) tailored for humanoid robotics control. The startup says it’s the first VLA to enable high-rate control of the entire upper body, including the wrist, fingers, torso, and head. With under 500 hours of training, robots powered by Helix can pick up virtually any object by voice command.
The AI model is also notable for its ability to run on multiple robots simultaneously. Figure previously shared demo footage of two Figure 02 robots collaborating to put away groceries after being given vague instructions. The company plans to offer its humanoids to the consumer market once their capabilities are refined in industrial settings.
Figure says the technology, which runs entirely on embedded low-power GPUs, is essential for bringing AI-powered humanoid robots into households at scale. The company proposes that combining intelligence collected from robot behavior with the Helix model will trigger exponential growth.
Sunnyvale-based Figure is developing its next-generation robot, presumably called the Figure 03, that’s optimized for manufacturing at scale. Adcock recently said Figure has signed on its second commercial customer, without naming the company. He said he sees a path for shipping up to 100,000 humanoid robots over the next four years.
Figure is seen as perhaps the biggest threat to Tesla, which plans to produce up to 10,000 Optimus humanoid robots this year. While Tesla’s production target is much larger than Figure’s, Elon Musk has admitted it’s highly aspirational.
Figure and Tesla face stiffening competition, with Google investing heavily in Apptronik and tech giants like Apple, OpenAI, and Meta reportedly planning to enter the market.