A new video released by Figure showcases their latest development featuring an undercover “actor,” Helix 02, which is a Vision-Language-Action policy-based control system guiding their humanoid robots. The footage depicts two robots, F.03, efficiently organizing a bedroom in under two minutes. The standout moment in the scene is when the robots work together on making the bed: positioned on opposite sides, they lift and straighten the duvet, adjust the creases, and collaborate on manipulating the flexible object. Figure explains that the two robots operate using a single trained network capable of translating images and instructions into actions. There is no shared planner, message exchange, or central control. Each robot observes the room through its own cameras and infers the other’s intentions from their movements.
The sequence also showcases a variety of other tasks. Through the robots’ limbs, Helix 02 is shown opening doors, hanging up a garment, placing headphones on a stand, closing a book, disposing of trash using the trash can pedal, and pushing a chair under the desk. These actions require walking, balance, dexterity, and continuous environmental awareness. However, the most intriguing aspect remains the collaboration around making the bed. A duvet lacks a fixed pose, it wrinkles, slips, and changes shape with each adjustment. When one robot alters the fabric tension, the other’s task changes instantly. While the demonstration is impressive, it focuses on a specific room layout, leaving uncertainties about the number of attempts needed and how the outcome varies with different object placements or furniture configurations not present in the training data.
In essence, the question remains whether the F.03 robots with Helix 02 could effectively operate in any other environment. The video raises fascinating possibilities for the future of robotic technology and collaborative intelligence.
About the Author
