US startup Figure's humanoid robot walks naturally like a human

The company said Figure 02 achieved its natural gait using reinforcement learning (RL), an AI method where controllers improve through trial and error based on a reward system, according to Interesting Engineering.

This allowed the robots to master heel strikes, toe-offs, and synchronized arm swings, replacing the rigid, mechanical movements of earlier models.

Figure said it trained its RL system in a high-fidelity simulator, compressing years of data into hours to help robots navigate real-world challenges like uneven terrain and sudden slips or shoves.

The video also shows ten Figure 02 robots walking using the same neural network without individual modifications, which the firm said highlights the scalability of its technology and the potential for mass deployment.

Beyond locomotion, Figure has integrated advanced AI models into its robots. In February, it introduced Helix, a Vision-Language-Action (VLA) system that combines perception, language understanding, and control to tackle key robotics challenges.

Figure 02 robots were also recently seen operating at BMW’s Plant Spartanburg in South Carolina, performing real-world tasks such as part retrieval and precision operations.

The deployment is part of BMW’s iFACTORY initiative, which aims to modernize car manufacturing through automation and smart technologies.

Figure projects 2025 as a pivotal year for growth, with plans to begin production, scale up shipments, and expand into home robotics.

The company is positioning itself as a major contender in the humanoid robotics sector, alongside Tesla’s Optimus, Agility Robotics’ Digit, and Chinese firms like UBTech Robotics and Unitree Robotics.