US's Figure AI humanoid robot completes longest fully autonomous task

In a demonstration released on Jan. 27, the robot, Helix 02, carried out 61 separate locomotion and manipulation actions in a full-sized kitchen, according to Figure AI.

The sequence included walking to the dishwasher, unloading dishes, navigating the room, placing items into cabinets, reloading the dishwasher, and starting it again.

Figure AI described the demonstration as the "longest horizon, most complex task completed autonomously by a humanoid robot to date."

The company said loco-manipulation, defined as the ability to move and manipulate objects as one continuous behavior, remains one of the most difficult problems in robotics. While humanoid robots have previously demonstrated abilities such as jumping, dancing, or moving objects, most systems rely on preplanned motions with limited real-time feedback. When conditions shift unexpectedly, performance often breaks down.

By contrast, Figure AI said Helix 02 is designed to continuously perceive, decide, and act, allowing it to walk while carrying objects, adjust its balance while reaching, and recover from errors in real time.

At the core of this capability is System 0, a learned whole-body controller trained on more than 1,000 hours of human motion data combined with sim-to-real reinforcement learning.

Rather than engineering separate reward functions, the system learns directly from human motion, enabling coordinated posture control and balance across a wide range of behaviors.

Helix 02 also integrates tactile sensing and palm-mounted cameras to enhance fine manipulation. In autonomous tests, the robot has unscrewed bottle caps, removed individual pills from organizers despite visual obstruction, dispensed precise syringe volumes under varying resistance, and selected small metal parts from cluttered containers, according to Interesting Engineering.

Figure AI said the results are still early but already show the potential of continuous, whole-body autonomy, adding that the work marks a shift away from choreographed demonstrations toward robots capable of genuine autonomous problem-solving.