Helix robots can understand and execute commands expressed in everyday language, possessing high-frequency, continuous control over the entire humanoid upper body.
Figure.ai, a tech company specializing in advanced robotic technologies with cutting-edge AI integrations, introduced Helix – a Vision-Language-Action (VLA) model for comprehensive humanoid robot control, allowing for dexterous manipulation, natural language interaction, and multi-robot collaboration.
This groundbreaking robotic product is capable of precise management of wrists, torso, head, and even individual fingers, enabling dexterous manipulation in real time. A unique feature of the new VLA model is that Helix can run simultaneously on two robots. That allows them to work together on complex, sophisticated tasks, e.g. collaboratively storing groceries. They may also perform certain tasks and manipulate daily objects by following natural language prompts. Impressively, the robot can identify and grasp a suitable object, even if it’s never encountered it before.
Unlike previous robotic systems that require task-specific fine-tuning, Helix uses a single set of neural network weights to learn a broad range of behaviors. This unified approach enables the robot to handle various tasks without separate training for each one.
According to the press release, Helix is designed to run on embedded, low-power GPUs, and is ready for real-world commercial deployment. Its efficient architecture, featuring a two-tier system (a slower, high-level vision-language module and a fast, reactive control module), ensures that it can operate smoothly in dynamic environments.
Except for Figure AI, Nvidia, Ant Group, Apptronik and Tesla have recently joined the race to deliver efficient humanoid robots. Reportedly, Meta Platforms is also planning to establish a new division within its Reality Labs unit to develop AI-powered humanoid robots that will help humans perform physical tasks.
