Why Humanoid Form Factor Is a Task Decision, Not a Look
The physical body a humanoid robot ships with decides which floors, tasks, and failure modes it can actually handle, independent of how capable its AI software is. A buyer or integrator comparing humanoid platforms should treat the leg-or-wheel decision the same way they would treat a forklift's tire choice: a mechanical constraint that limits the job, not a cosmetic detail.
Three form factors currently cover almost every commercial and research humanoid on the market: fully bipedal robots that walk on two legs, wheeled humanoids that put a human-like torso and arms on a rolling base, and wheel-legged hybrids that try to combine both. Each solves a different piece of the same underlying problem: operating in spaces built for human bodies without redesigning the building around the robot.
Bipedal Humanoid Robots
A bipedal humanoid walks on two legs and is built to move through stairs, curbs, and uneven or cluttered terrain without any changes to the space. Current examples include Boston Dynamics' Atlas, doing sequencing work at Hyundai's Metaplant in Georgia; Tesla's Optimus Gen 3, built noticeably lighter than Atlas to favor endurance and battery life over athletic performance; Agility Robotics' Digit, which has moved more than 100,000 totes at a GXO warehouse; and Figure AI's Figure 03.
The tradeoff is mechanical and controls complexity. A biped is dynamically unstable by design and needs continuous real-time balance correction, which adds joint wear, fall risk, and power draw that a wheeled base never has to solve. Walking is also slower and less energy-efficient than rolling on the same floor, and every added degree of freedom for dexterity is another joint that can fail.
Bipeds are currently deployed mostly in manufacturing and warehouse pilots (auto plants, logistics floors) plus ongoing R&D and data-collection programs, rather than in the varied indoor-outdoor terrain their legs are built for. A fair criticism from the robotics community is that most public bipedal demos still happen on flat factory floors, so the stair-and-rough-terrain case for legs remains mostly theoretical in production settings today.
Wheeled Humanoids
A wheeled humanoid keeps a human-like torso, arms, and hands but replaces the legs with a wheeled mobile base. Current examples include Sanctuary AI's Phoenix, piloted at Mark's and Canadian Tire locations and partnered with Magna; Galbot's G1, an omnidirectional wheeled platform demonstrated at China's 2026 CCTV Gala; and Diligent Robotics' Moxi, a wheeled hospital-delivery robot deployed across more than 25 U.S. hospitals.
Wheels remove the need for active balance control entirely, which buys higher speed, lower mechanical cost, and much longer uptime between failures than a biped can currently offer. Sanctuary AI has said publicly that its move to a wheeled base followed direct customer feedback that legs were too mechanically frail to support a torso doing precise, forceful work reliably.
The cost is absolute: a wheeled humanoid cannot climb stairs and needs flat or ramped floors, a real constraint in older buildings or unfinished construction sites. That limitation is structural, not a solvable software problem, which is why wheeled platforms concentrate in hospitals, retail floors, and warehouse environments where the operator already knows the floor is flat. Critics sometimes argue a wheeled base isn't meaningfully humanoid in its locomotion and gives up any claim to drop-in-replace a human worker in a space with steps or thresholds.
Wheel-Legged Hybrid Humanoids
A wheel-legged hybrid tries to combine both approaches, typically by ending each leg in a wheel instead of a foot so the robot can roll on flat ground and switch to stepping when the floor demands it. Current examples include GAC Group's GoMate, targeting small-scale production in 2026, and Hexagon Robotics' AEON, piloted at a BMW plant in Germany since 2025.
The stated goal is stair-and-rough-terrain capability without giving up wheeled speed and efficiency on flat runs. That combination is not free: published research on transformable wheel-leg designs shows the mechanism can cost a meaningful share of a pure-legged design's workspace volume, and switching between rolling and stepping modes requires added control sophistication to manage the robot's center of mass during the transition.
Hybrids currently target automotive manufacturing floors that mix flat runs with ramps and obstacles, which is exactly the environment where neither a pure biped nor a pure wheeled base is a clean fit. This remains the least commercially proven of the three categories: most public examples are pilots or research platforms, so real-world reliability and total cost of ownership data are still thin compared to established wheeled and bipedal deployments.
| Form factor | Best for | Cannot do | Example |
|---|---|---|---|
| Bipedal | Stairs, curbs, human-built spaces with no modification | Match a wheeled base's speed, efficiency, or uptime | Boston Dynamics Atlas, Tesla Optimus |
| Wheeled | Flat or ramped floors needing speed, stability, long uptime | Climb stairs or cross unramped steps at all | Sanctuary AI Phoenix, Diligent Moxi |
| Wheel-legged hybrid | Mixed flat-and-obstacle floors (e.g. factory floors with ramps) | Match a pure biped's leg workspace or a pure wheeled base's mechanical simplicity | Hexagon AEON, GAC GoMate |
Choosing a Form Factor for Your Own Deployment
Start with the floor, not the robot. Walk the actual deployment space and note every stair, curb, threshold, and unramped step the robot would need to cross unattended. If that list is empty, a wheeled humanoid gets more uptime and lower cost for the same task than a biped that pays a mechanical-complexity tax it never needs to spend.
Weigh maturity against capability. Wheeled platforms in hospitals and retail have the longest deployment track records; bipeds are still concentrated in single-site manufacturing pilots; wheel-legged hybrids have the shortest public deployment history of the three. A buyer who needs proven uptime today should weight that history at least as heavily as a spec sheet's top-line capability claim.
- Map every stair, curb, and threshold in the deployment space before comparing platforms.
- If the floor is entirely flat or ramped, a wheeled humanoid will likely out-perform a biped on uptime and cost.
- If the floor mixes flat runs with steps, weigh a wheel-legged hybrid's added complexity against simply routing around the obstacle.
- Ask any vendor for deployment count and duration, not just a demo video — bipedal and hybrid demos in particular tend to run on flatter, more controlled floors than their marketing implies.
Bottom Line
Bipedal, wheeled, and wheel-legged hybrid humanoids are not three versions of the same robot; they are three different mechanical bets on which floors and tasks matter most. Bipeds trade cost and uptime for stair and rough-terrain access that most current deployments don't yet use. Wheeled humanoids trade stair access entirely for speed, stability, and the longest deployment track record of the three. Hybrids bet on getting both at the cost of mechanical complexity and the thinnest real-world reliability data. Match the form factor to the actual floor plan, not the demo video.
Walk your deployment floor and count the stairs and thresholds before comparing humanoid platforms on spec sheets alone.
FAQs
Why do some humanoid robots use wheels instead of legs?
Wheeled bases remove the need for active balance control, which lowers mechanical cost and complexity and increases uptime and speed compared to a bipedal design. Sanctuary AI has said its move to wheels followed customer feedback that legs were too mechanically frail for precise, forceful work. The tradeoff is that a wheeled humanoid cannot climb stairs at all.
Are bipedal humanoid robots actually used on stairs and rough terrain today?
Rarely in current commercial deployments. Most public bipedal pilots, including Boston Dynamics' Atlas at Hyundai and Agility's Digit at GXO, run primarily on flat manufacturing or warehouse floors. The stair-and-rough-terrain case for legs is the stated long-term justification for the design, not yet the typical day-to-day task.
What is a wheel-legged hybrid humanoid robot?
A design where each leg ends in a wheel instead of a foot, letting the robot roll on flat ground and switch to stepping when it meets an obstacle. Hexagon Robotics' AEON, piloted at BMW, and GAC Group's GoMate are current examples. The approach adds mechanical and control complexity in exchange for handling both flat and uneven floors with one platform.
Which humanoid form factor has the longest deployment track record?
Wheeled humanoids. Diligent Robotics' Moxi has been deployed across more than 25 U.S. hospitals, giving the wheeled category the longest and broadest real-world operating history of the three form factors.
How should I choose between a bipedal and a wheeled humanoid for a warehouse or factory?
Walk the actual floor first and list every stair, curb, and unramped threshold the robot would need to cross unattended. If that list is empty, a wheeled humanoid will typically deliver more uptime and lower cost for the same task. If the floor has real obstacles a wheeled base cannot cross, a bipedal or wheel-legged hybrid platform becomes the realistic option, budget and maturity permitting.