Build vs Buy a Research Humanoid: Cost and Time

Every capable robotics lab eventually asks it: why buy a humanoid when we could build our own? It is a reasonable instinct — you have the talent, and a custom robot sounds cheaper and more flexible. But the build-versus-buy math is rarely what it looks like at first. This guide walks through the real cost and timeline of each path so you can decide with eyes open.

The true cost of building

A research humanoid is not one project; it is a dozen, each demanding specialists:

• Mechanical design — kinematics, structure, actuator selection, iteration through physical prototypes.
• Actuators and electronics — motors, drivers, power, wiring, thermal management.
• Firmware and low-level control — real-time joint control, safety, calibration.
• Perception and compute — cameras, synchronization, on-board hardware.
• Software — an SDK, a URDF model, and a simulator, plus a teleoperation and data pipeline if you want to do imitation learning.
• Maintenance forever — spare parts, repairs, and keeping the whole stack working.

Add it up and a credible in-house humanoid is measured in person-years, not person-months — and that is before a single research experiment runs on it.

The cost that hurts most: time and focus

The line item that rarely makes the spreadsheet is opportunity cost. While your team designs actuators and debugs firmware, it is not doing the research you hired it to do. A typical custom build puts a useful robot 1–3 years out. For a lab whose goal is manipulation, imitation learning, or VLA research, that is years of your actual mission spent building a tool instead of using one.

A realistic build timeline

Teams routinely underestimate this because they price the first prototype, not a working research tool. A more honest arc looks like:

• Months 1–6: mechanical design, actuator selection, first prototype that moves.
• Months 6–15: firmware, safety, calibration, perception, and the painful integration where subsystems meet.
• Months 12–24+: the software a researcher actually needs — SDK, URDF, simulator, teleoperation — without which the hardware cannot do imitation-learning work.
• Forever: maintenance, spares, and keeping the stack alive as people rotate through the lab.

And version one is rarely the robot you want — most real designs need a second iteration before they are dependable.

The maintenance tail

Buying ends a build project; it does not end a maintenance project. A custom humanoid needs someone who understands the whole stack on call for years — and when that person graduates or leaves, undocumented tribal knowledge walks out with them. Bus factor is a real, recurring cost of the build path that spreadsheets almost never capture.

When building is the right call

Building genuinely makes sense when:

• Your research is the hardware — novel morphologies, new actuators, exotic kinematics.
• You are a dedicated hardware lab with the staff and budget to sustain it for years.
• No existing platform can express the thing you are trying to study.

When buying wins

Buying wins when your goal is to do research with a humanoid rather than to produce one:

• You want to collect data and train policies this quarter, not in two years.
• You want your specialists working on ML and manipulation, not motor drivers.
• You still want room to customize — good platforms are hackable, so you are not locked out of modifying them.

Where Prometheus fits

Prometheus is the buy that keeps the upsides of building:

• Productive on day one — SDK, URDF, simulator, cameras, and a teleoperation pipeline included, so you start collecting demonstrations immediately. See how to collect data and fine-tune a VLA.
• Still hackable — open architecture and modular hardware let you swap arms, grippers, and base modules and write your own controllers. You customize where it matters without building the other 90%.
• Supported — direct engineering help instead of a maintenance burden you carry alone, and it is designed and made in the EU.

If the decision really comes down to budget, our breakdown of what a humanoid robot costs puts the buy option next to the very real, often-underestimated cost of building.