Human-like Gait (Straight-knee, Heel-to-toe Walking)

Having a humanoid robot walk the way people do — knee straight, heel touching down first, then pushing off with the toe.

‘Human-like gait’ is common usage in the humanoid robotics industry for a walking style close to a person's: the stance leg's knee stays essentially straight, the heel lands first, the weight rolls across the foot, and the toe pushes off at the end, with the arms swinging opposite the legs. Early zero-moment-point (ZMP)-based humanoids mostly walked with bent knees, flat feet, and constant hip height, partly because a straight knee sits close to a kinematic singularity, making hip-height control difficult. Waseda University's WABIAN-2R, in 2006, was an early success at straight-knee, heel-to-toe walking, using hip motion to avoid the singularity and adding a passive toe joint to the foot. In recent years this has mostly been achieved with reinforcement learning: a reward term imitating a human walking reference trajectory is added, balanced against velocity tracking and energy cost.

ExampleIn March 2025, Figure released a reinforcement-learning walking controller for Figure 02: rewarded in GPU simulation for imitating a human walking reference trajectory, it learned heel strike, toe-off, and arm swing synchronized with the legs, and transferred zero-shot to the real robot. China's EngineAI launched its SE01 in October 2024, reportedly marketed with straight-knee gait as a key selling point.

Also called

Human-like Walking, Heel-Strike / Toe-Off Walking

Related

Straight-Knee Walking · Gait · Zero Moment Point · Bipedal Locomotion · Adversarial Motion Priors · Stance Phase / Swing Phase

Sources

Ogura et al.: Human-like Walking with Knee Stretched, Heel-contact and Toe-off Motion by a Humanoid Robot (IROS 2006)
Figure: Natural Humanoid Walk Using Reinforcement Learning (2025-03)
科创板日报：深圳人形机器人行走视频走红 拟人步态震惊英伟达科学家（2025-01） (Chinese)

As of

2025-03