Why We're Putting Unitree A2-W at the Centre of Our Construction Survey Stack

Most robotics procurement decisions for construction start in the wrong place. People look at the robot first, argue about IP54 ratings and gait algorithms, then get surprised when a six-figure platform has eaten the entire budget before a single sensor is bolted on.

We've been approaching it the other way around.

When we spec'd out our surveying solution for construction sites, we started with the sensors: the LiDAR unit, the GNSS receiver, the photogrammetry payload. Those are where the precision actually comes from. The robot underneath is a carrier, a mobile tripod that needs to be durable, programmable, and affordable enough that the budget math works. Unitree's A2-W fits that description better than anything else we've evaluated.

What Unitree Actually Is

Unitree is a Chinese robotics manufacturer that has spent the better part of a decade doing to quadruped robots what BYD did to electric vehicles: systematic cost reduction without gutting the core engineering. Their Go2 quadruped now retails for around $1,600. The humanoid G1 sits at roughly $27,000 to $30,000, shipping in volume while Western competitors are still in prototype rounds. SemiAnalysis published a detailed breakdown of their trajectory earlier this year, noting that Unitree is approaching 10,000 humanoid units shipped with at least 250 already in active industrial deployments (https://newsletter.semianalysis.com/p/chinas-unitree-will-dominate-global).

The A2-W sits in a different tier: a high-performance quadruped with wheeled feet rather than purely articulated legs. That distinction matters enormously for construction applications.

Why Wheeled Feet Change the Equation

Legged robots handle rough terrain well. But construction sites, particularly during mid-build and finishing phases, have long stretches of compacted hardcore, concrete screed, and flat slab that are navigationally trivial. A purely legged robot covers those sections slowly and wastes battery doing it.

The A2-W's hybrid approach gives you the obstacle-crossing capability of a quadruped when the terrain demands it, and smooth, fast rolling on flat surfaces when it doesn't. In practice, that means better range per charge, more consistent survey coverage per shift, and fewer interruptions. Fluidity of movement also matters for certain sensor payloads: vibration from a stilted gait degrades point-cloud quality in ways that are hard to correct in post-processing.

The Sensor Stack Is Where the Real Investment Goes

A construction survey is only as good as its data. The A2-W's job is to get the sensor to the right place, repeatably, without falling over. The actual precision work is handled by:

• LiDAR (we're currently evaluating units in the Ouster OS0/OS1 range and Livox Mid-360 for close-quarters internal surveys)
• GNSS with RTK correction for georeferenced external scans
• Photogrammetry cameras where point-cloud density needs supplementing with texture

This is where we're putting the bulk of the project budget. A high-quality LiDAR payload for construction-grade survey work costs more than the robot itself. That inversion is the point. The robot being affordable means we can spend properly on the instruments that actually determine whether the output is usable by a structural engineer or a quantity surveyor.

What We Think

Unitree's pricing isn't a signal of cut corners. Their actuator manufacturing is now mature enough that the cost reductions reflect volume and vertical integration, not compromised components. For construction use cases, you don't need a Boston Dynamics Spot. You need a platform that is tough enough to survive a busy site, open enough to integrate custom sensor payloads, and cheap enough that deploying two units on a large project doesn't require a board-level sign-off.

The A2-W checks those boxes. So we've made our choice.

What to Watch

The next question is software. Raw scan data from a mobile platform only becomes useful when it's processed, compared against BIM models, and surfaced in a way that site managers can act on. That's where the harder integration work sits, and it's the layer we're building now.