Figure Demonstrates Humanoid Robot Walking Naturally Using Reinforcement Learning
Figure introduced an end-to-end neural network trained with reinforcement learning for humanoid locomotion. The training involved thousands of virtual robots in a high-fidelity physics simulator, compressing years of data into hours. The learned walking model transferred directly to real-world Figure 02 robots without additional tuning.
Evidence notes
- Figure introduced an end-to-end neural network trained with reinforcement learning for humanoid locomotion.
- The training involved thousands of virtual robots in a high-fidelity physics simulator, compressing years of data into hours.
- The learned walking model transferred directly to real-world Figure 02 robots without additional tuning.
Company context
Figure AI develops humanoid robots and autonomy systems with a focus on human-like reasoning and speech-to-action control.