A company called Figure built Helix, a new tool for humanoid robots, TechCrunch reports. Helix is a Vision-Language-Action model, or VLA. A VLA is a system that uses sight and words to guide robots.

VLAs are new in robotics. They mix vision and language to process info. Google’s RT-2 is a famous one. It teaches robots using videos and big language models (LLMs). Helix takes visual clues and spoken commands to move a robot instantly.

Figure says Helix can handle tons of new household stuff. Think of cups, toys, or spoons it’s never seen. You just tell it what to do in plain words. Like, “Pick up the red mug.” The robot looks around, spots it, and grabs it. Figure wants robots to obey simple orders easily. Helix helps by linking what the robot sees with what it hears.

Robots struggle in homes because learning and control are tough. Plus, they cost a fortune - tens or hundreds of thousands of dollars. Most robot companies focus on factories first. They want to make robots trusty and cheap before homes.

Homes are good testing grounds for smart robots

But Figure says homes matter a lot. They’re tricky spots to test smart robots. Teaching them kitchen tasks opens doors to other jobs. Figure explains robots need to learn new tricks fast, especially with strange objects. Old ways, like coding by hand, take too long. Experts spend hours programming one task. Or robots need thousands of practice runs.

Homes are wild - kitchens differ, tools vary, messes happen. Helix aims to fix that.

Figure announced Helix on X, with a video. A longer video is on YouTube. Figure also shared a report titled "Helix: A Vision-Language-Action Model for Generalist Humanoid Control."

It’s early days, though. The cool videos hide tons of work. Figure wants to hire more brainy folks to grow Helix.

"While these early results are truly exciting, we think they only scratch the surface of what is possible," concludes the report. "We are eager to see what happens when we scale Helix by 1,000x and beyond."