Figure 03: Humanoid robot autonomously climbs a 'ladder'
Figure
Figure AI has taught its humanoid Figure 03 robot to autonomously climb a ladder, or more precisely a ladder staircase. The robot uses the Vision-Language-Action model 'Helix' to perform complex mobility tasks with visual self-monitoring. The achievement is notable as ladder climbing is considered a pinnacle of humanoid robot locomotion control.
US robotics company Figure AI published a video on X showing its humanoid Figure 03 robot autonomously climbing a three-rung ladder staircase onto a platform. Figure AI founder Brett Adcock stated that the F.03 can now climb a ladder fully autonomously, without teleoperation. In the video, the robot repeatedly analyzes its surroundings to find handholds, slowly pulling itself up. The exact success rate is undisclosed. The control and coordination are likely handled by the Vision-Language-Action model 'Helix', which runs locally on Figure 03 robots. 'Helix' recently received an update combining visual perception with whole-body motion control, enabling autonomous navigation in unknown spaces. The current version incorporates images from stereo cameras to perceive the environment and its own body posture, improving precision, balance, and fluidity in challenging terrain. Climbing a ladder is considered one of the most demanding tasks in humanoid robot motion control, requiring precise perception, coordination of limbs and torso, and real-time balance. Since it is a ladder staircase with a fixed railing, the task is easier than climbing a steep ladder. For comparison, ETH Zurich's four-legged robot ANYMal learned to climb a vertical ladder via reinforcement learning in 2024, achieving a 90 percent success rate.
- Abbreviations
- ETH = Eidgenössische Technische Hochschule — Швейцарская высшая техническая школа Цюриха
- VLA = Vision-Language-Action — модель «зрение-язык-действие»
Source: Heise online —
original
