Legged locomotion

What it means

Legged locomotion moves a body by making and breaking supporting contacts through its legs. Walking and running involve different contact sequences and dynamics, so the general term covers more than a single gait. MIT's simple-leg models use stance and swing motion, passive walkers and simplified running dynamics to explain the basic mechanics. [1]

Locomotion is also distinct from navigation. Executing a gait does not establish that a robot independently chooses a route or understands a destination. Humanoid-Gym reports locomotion-skill training and transfer; those evaluated tasks should not be expanded into an unsourced claim about general navigation. [2]

How it applies to humanoids

A biped has to coordinate support and body motion as its feet change contact. Simple models can reveal important behavior before adding all the details of an articulated humanoid. MIT's passive-walker example demonstrates a stable repeated gait without motors or a controller by descending a small ramp under gravity. It illustrates how mechanics and the energy supplied by the environment can contribute to walking. [1]

Learning is one approach to generating humanoid locomotion. Humanoid-Gym reports simulation training and physical transfer on two RobotEra platforms. Unitree describes G1 as imitation- and reinforcement-learning driven and lists its leg joint counts separately from arms and hands. These sources provide different evidence: a research evaluation and a manufacturer platform description. [2, 3]

Gait and measurement traps

A walking speed, running speed and one-off sprint result should retain their gait and test conditions. Locomotion research evaluates particular behaviors under particular environments; a result in one setup does not establish stairs, uneven ground or loaded carrying unless those were tested. [1, 2]

Simulation success is not physical success. Humanoid-Gym explicitly separates cross-simulator verification from real-world transfer. A successful simulated gait should therefore retain its environment and evaluation scope, rather than silently becoming a verified hardware capability or a claim that all robot configurations behave alike. [2]

Stable repetition and disturbance recovery are different questions. MIT's passive-walking discussion explains stability of a repeated gait within a model. That does not mean every disturbance, contact surface or payload is covered. Read the experiment's assumptions and perturbations before interpreting stable walking as broad robustness. [1]

How to read it on humanoidversus.com

Use form factor to identify bipeds, then read max_speed_mps and its note for the sourced motion being described. payload_kg and runtime_h provide separate operating claims; do not assume they were measured at the listed maximum speed or during the same gait.

G1 is linked because its record cites Unitree's product specifications. This entry has no live ranking. Peak torque and joint counts describe hardware context, while walking reliability requires task-specific evidence. A null speed or unknown gait should remain unknown rather than be inferred from a short clip. [3]

Further reading

MIT's simple-leg chapter explains walking and running mechanics. Humanoid-Gym documents a humanoid learning and transfer framework. Unitree's G1 page illustrates how a manufacturer's hardware specification and learning description provide context without supplying a standardized locomotion score. [1–3]