American robots walked as flexible as people

In the video, the Figure 02 -shaped robot has a flexible step, naturally instead of rigid gait, machines like many other robots. Figure 02 uses enhanced learning technology in a physical simulation process, simulating many years of data in just a few hours. The random process of domain allows smooth transition from simulation to walking in the real world.

Enhancing learning (RL) is a method of applying AI, in which the controller learns through the test and the wrong process, allowing Figure 02 to walk with human gait. Figure has trained its RL controller in highly honest GPU accelerating physics simulation. Thousands of virtual people -shaped robots, each robot has different physical parameters, is operated in parallel.

This simulation helps the robot come into contact with many realistic situations such as diverse terrain, transmission dynamics and challenges such as sliding, stumbling, pushing. Figure AI asserts that, through this extensive learning environment, a unique artificial neurological network policy is developed to manage robots’ movements.

A major advantage of Figure AI approach is the ability to convert this policy directly from simulation to real robots without further adjustment. The process of seamless conversion ensures the ability to walk skillfully like people in many different environments. The RL -based training method significantly reduces the development cycle, while improving the adaptability and reliability of the robot.

Figure 02 can currently perform human -like movements including grounding with heels, toes and swinging hands synchronously. These improvements are achieved by using RL training process, “rewarding” for robots when it imitates human walking, while optimizing velocity, energy efficiency and durability.

In the video, Figure AI introduces 10 Figure 02 robots to work on the same RL artificial nerve network without modification. This homogeneity shows the potential to increase technological scale, opening the ability to deploy thousands of robots without manual adjustment.

Figure Ai who is locating their robots is a competitor in the field of human -shaped robot with Tesla’s Optimus robot, Agility Robotics and robot of Chinese companies such as Ubtech Robotics, Unitree Robotics. 2025 is expected to be a hinge year with Figure AI when the company started to focus on production, increased the amount of shipped robots and made advances in family robots.