FigureAI has developed a new AI-powered walking controller for its Figure 02 humanoid robot, taking the android movement patterns into slightly more natural territory.
Announced yesterday, the update was showcased in new footage comparing the previous and current movement systems. There's a clear difference - while the robot's gait still appears mechanical, it now includes more human-like elements such as heel strikes, toe-offs, and synchronized arm swings. It's a subtle improvement, but enough to push the Figure 02 away from the rigid, protocol-droid motion of earlier iterations.
The system was trained entirely in simulation using reinforcement learning - a method where virtual robots learn to walk through trial and error. That training was then transferred directly to the real-world robot without manual tuning, allowing for consistent performance across multiple machines.
Figure is one of several startups chasing the sci-fi-driven vision of general-purpose, humanoid robots, with domestic competitors like Tesla's Optimus, and Agility Robotics' Digital also pushing into the space. Chinese firms have also been determined to deliver a large fleet of humanoids by the end of this year, and recently, NVIDIA announced a significant push into robotics with its GR00T N1 model, and its Omniverse platform.
The Figure 02 is primarily designed for industrial and manufacturing purposes, lending a hand in reducing labour shortages, and reducing the number of workers in unsafe jobs. However, there are countless ventures aiming to merge humanoids into everyday homes. Regardless of whether they're in the home or the warehouse, humanoid robotics are required to interface with humans effectively, both in terms of movement patterns, communication, and the ability to carry out complex functions.
Recently, Figure showcased its Helix model, which enables humanoids to see, understand and interact with the world in real-time. These advancements are just one of many, in a space that continues to see new innovations like this every day.