US-based humanoid startup Figure AI has announced Helix 02, its updated humanoid robotics model, following the first Helix version announced early last year.
This enables a single neural system to control walking, manipulation and balance across the entire robot, using only onboard sensors and without human intervention.
The company said Helix 02 can complete a four-minute kitchen task that includes walking to a dishwasher, unloading dishes, moving across a room, stacking items, reloading the dishwasher and starting it. Figure said the task ran end-to-end with “no resets and no human intervention”, marking its longest autonomous demonstration to date.
The robot also uses its full body during tasks. In the kitchen demonstration, it closed a drawer with its hip and lifted a dishwasher door with its foot when its hands were occupied, according to the company.
Figure said Helix 02 executed 61 ordered actions during the dishwasher task and maintained the task context for several minutes, with implicit recovery from minor errors.
Helix 02 extends Figure’s earlier Helix system, which controlled only the upper body, to full-body autonomy. According to the company, the new model connects vision, touch, and proprioception sensors directly to all actuators via a single visuomotor neural network.
At the core of the system is what Figure calls System 0, a learned whole-body controller trained on more than 1,000 hours of human motion data. Figure said System 0 replaces over one lakh lines of hand-written control code with a single neural model that handles balance and coordination at high frequency.
The system operates alongside two other layers.
System 2 handles scene understanding and language-level goals, while System 1 translates sensor inputs into joint-level commands for the entire body. Together, the three layers allow the robot to perceive, decide and act continuously, rather than switching between separate walking and manipulation modes.
The company added that this approach avoids the traditional use of state machines, where robots must stop walking before reaching or grasping objects. “True autonomy requires something fundamentally different: a single learning system that reasons over the whole body at once,” the company said.
In addition to room-scale tasks, Helix 02 supports fine manipulation using tactile sensors and palm-mounted cameras embedded in Figure’s latest hardware platform. These sensors allow the robot to operate when objects are occluded from its main cameras and to regulate grip force during contact.
Figure demonstrated tasks such as folding laundry, unscrewing bottle caps, extracting a single pill from a medicine organiser, dispensing a precise volume from a syringe and picking small metal parts from clutter. The company said these tasks rely on touch feedback and in-hand vision, rather than vision alone.
The company described the results as early but said they point towards humanoid robots that can operate autonomously over extended periods in real-world environments, combining locomotion and manipulation as a single, continuous behaviour.
ALSO READ: NVIDIA Unveils Open Earth-2 Models for AI Weather Forecasting
