Figure says its Helix 02 humanoid robot performed a range of living-room tidying behaviours using a single general-purpose neural system. In a company-produced demonstration published on March 9, the robot walked through a room, used cleaning tools, handled rigid and flexible objects, carried containers, operated a television remote and moved through a narrow gap between furniture.

The demonstration extends Figure’s previously described Helix 02 task set from kitchen cleanup to a less structured household environment. It is notable for combining locomotion and manipulation rather than showing the robot performing every action from a stationary position.

Figure has not provided quantitative results for the demonstration. The supplied material does not establish the task-completion rate, number of attempts, repeatability, duration, failure modes or level of human supervision.

Contents

What Helix 02 demonstrated

Figure describes Helix 02 as a system that controls the robot’s full body directly from camera pixels. This means, in Figure’s description, that visual input is used to control the robot’s full-body behaviour.

The living-room demonstration included several distinct behaviours:

  • The robot used a spray bottle to wet a dirty surface and a towel to wipe it with forceful motions.
  • It unhooked a towel from its arm, repositioned the towel for cleaning and later threw it over its shoulder to free its hands.
  • It picked up a bin with both hands and scooped blocks from a table into it.
  • It tucked a container under one arm, leaving both hands free to collect toys.
  • It threw a pillow onto a couch using what Figure describes as a fast, controlled motion.
  • It picked up a remote, reoriented it in its hand and pressed the correct button to turn off a television.
  • It temporarily stowed a towel under one arm while moving between tasks.
  • It side-stepped through the narrow space between a coffee table and a couch while continuing to manipulate an object.

These examples are technically different. A spray bottle and towel require coordinated tool use. A towel is a flexible object whose shape changes during handling, making it harder to model and control than a rigid object. Picking up and scooping with a bin requires both hands to stabilise and manipulate the container. Reorienting the remote requires the robot to change the object’s pose while maintaining grasp control.

The examples also require the robot to reorganise its body and carried objects as the task changes. Stowing a towel, carrying a container under one arm and continuing to walk while manipulating an object are whole-body strategies rather than isolated hand movements.

The technical significance

Whole-body robot control coordinates locomotion, balance, arm and hand movement, and sensing. This matters in a home because useful actions can require the robot to move around furniture while maintaining control of an object or tool.

Compared with a fixed industrial workstation, a living room can have changing object locations, furniture layouts and object behaviour. The robot therefore faces the combined challenge of positioning its body, maintaining balance and handling objects as it moves through the room.

The demonstration stresses transitions between actions. The robot is shown moving from cleaning to handling a towel, from carrying a container to picking up toys, and from navigating around furniture to manipulating an object. The technical challenge is not simply adding a list of individual skills; it is coordinating them as conditions change.

Figure says these behaviours were learned by adding new data, without new algorithms or task-specific engineering. It also says the same general-purpose architecture used for previous tasks controls the behaviours in this demonstration.

If such an approach generalises beyond the demonstration, a shared control architecture could reduce the amount of special-case engineering required for different household tasks. The supplied evidence does not yet show that this generalisation has occurred reliably.

What the demonstration does not establish

This is a company-produced robotics demonstration, not a peer-reviewed study or an independently evaluated benchmark. Figure provides qualitative examples but no formal experimental design, comparator or sample size.

The source does not report:

  • task-completion or success rates;
  • the number of attempts;
  • execution speed or demonstration duration;
  • repeatability across repeated trials;
  • failure and recovery behaviour;
  • safety incidents or collision rates;
  • the amount and type of training data;
  • whether a person intervened or provided remote assistance.

It is also unclear whether Helix 02 operated fully autonomously throughout the demonstration. The source does not specify the exact model version, hardware configuration or neural-system architecture beyond the general description of direct control from camera input.

A successful run in one reported environment does not establish reliable operation in ordinary homes. Performance may change with different room layouts, lighting conditions, object types, clutter levels and furniture arrangements. The demonstration also does not show that the robot is available for home purchase, suitable for unsupervised domestic use or ready for continuous operation around people.

What to watch next

The most useful follow-up evidence would be quantitative. Figure could report how often the robot completes each task, how many interventions are required, how long actions take and how performance changes across repeated trials.

It would also be important to test the system across different homes or room layouts, lighting conditions and object types. Flexible objects such as towels, for example, can behave differently depending on how they are placed and grasped. Narrow-space navigation should be assessed alongside collision avoidance and recovery from mistakes.

Further demonstrations may clarify how much of the environment must be prearranged and whether a person monitors the robot. Availability, pricing, safety information and real-world deployment have not been announced in the supplied material.

For now, Helix 02’s living-room demonstration is evidence that Figure has produced a demonstration combining walking, tool use, bimanual manipulation, object reorientation and household navigation within one reported control architecture. It is not evidence of measured reliability or unsupervised household autonomy.

Sources