Humanoid Robotics Technology
Figure AI has demonstrated a new capability of its Helix 02 system by showing a humanoid robot autonomously tidying a living room. The demonstration follows the company’s earlier release of Helix 02, a single neural system designed to control a robot’s entire body directly from visual inputs, enabling dexterous manipulation and long-horizon autonomy across an entire environment.
While “tidy the living room” may appear to be a simple household task, it presents a number of challenges from a robotics perspective. Living rooms are highly unstructured environments where objects are scattered unpredictably, furniture creates narrow navigation paths, and soft items such as towels or pillows behave dynamically. Many actions require coordinated two-handed manipulation, while others require the robot to free one hand during a task. Nearly all behaviors involve moving through the room while manipulating objects simultaneously.
In the demonstration, a humanoid robot powered by Helix 02 performs an end-to-end living room cleanup sequence. The robot walks through the room while continuously interacting with objects, tools, and containers, demonstrating coordinated whole-body autonomy that combines locomotion, dexterity, and environmental awareness.
The robot is able to perform coordinated tool use by operating a spray bottle to wet a surface before wiping it clean with a towel. It also manages flexible objects dynamically, repositioning a towel for cleaning and storing it temporarily to free its hands. Complex bimanual manipulation is demonstrated when the robot lifts a bin with both hands and then collects blocks from a table into the container.
Additional behaviors include using whole-body strategies for efficiency, such as tucking a container under one arm while picking up toys with both hands. The robot also performs dynamic actions such as tossing a pillow back onto a couch with controlled motion. Fine manipulation tasks are shown as well, including picking up a remote control, reorienting it within the hand, and pressing the correct button to turn off a television.
Throughout the task, the robot reorganizes tools and objects while moving between activities. It can temporarily store items under one arm while transitioning between tasks and navigate tight spaces by precisely stepping through narrow gaps between furniture while continuing to manipulate objects.
According to Figure AI, these capabilities were achieved by expanding the system’s training data rather than introducing new algorithms or specialized controllers. Helix 02 learns new behaviors through additional data, allowing the same general-purpose architecture to perform increasingly complex tasks that integrate locomotion, dexterous manipulation, and real-time environmental interaction.
The company notes that tasks such as tidying a living room combine multiple robotic capabilities simultaneously, including navigation, manipulation, tool use, and task planning in a dynamic environment. Demonstrating these behaviors with a single unified model represents a step toward scalable humanoid intelligence, where one system can learn new capabilities simply by observing more examples and expanding its repertoire of real-world behaviors.
