Robot 101 · Chapter 10

From prototype to production: the manufacturing journey

Two different engineering problems

In the prototype phase — typically the first one to three units — the goal is not manufacturing, it is learning. Components are sourced from whichever supplier can deliver fastest, even at three to five times what the same part will cost in production. 3D printing often replaces machined metal for structural parts and brackets where the function allows it; off-the-shelf motors, development boards, and evaluation-kit sensors stand in for the custom electronics that will eventually be designed. It is common for the prototype bill of materials to run four to eight times the target production cost, and that is expected — the question at this stage is whether the robot performs the target tasks, whether the software stack runs on the hardware, and whether any physics surprises (thermal hotspots, resonant frequencies, torque limits) demand a design change before anyone commits money to tooling.

A 100-unit production run is a different problem entirely. Every unit has to meet the same tolerances, pass the same tests, and be traceable back to the specific batch of components it was built from. A part that "worked fine" on a hand-built prototype can become a production headache the moment it has to be made the same way one hundred times in a row by a factory floor, not a bench. That shift is why teams treat the move from prototype to volume as its own phase, with its own reviews and its own failure modes, rather than simply ordering more of the same parts.

Design for manufacture and design for assembly

Before committing to a tooled build of even ten units, most teams run a design-for-manufacturing (DFM) and design-for-assembly (DFA) review. DFM asks whether a part can be made repeatably, at the specified tolerances, by a standard process — CNC milling or turning, injection molding, sheet-metal forming — or whether it requires a specialized process such as Swiss turning, EDM, or multi-axis simultaneous milling that adds both cost and lead time. A part that looked simple in CAD sometimes turns out to require a process only a handful of shops can run economically, which quietly triples unit cost.

DFA asks a complementary question: how many steps does a trained technician need to install a given subassembly, and which of those steps are prone to error — small fasteners reached through deep recesses, asymmetric parts that can be installed backwards, wiring that has to be routed by feel rather than by a visible guide. Reducing part count, adding self-locating features, and making asymmetric parts either symmetric or impossible to install wrong are standard DFA moves that pay for themselves many times over once hundreds of technician-hours are on the line instead of one engineer's afternoon.

Qualifying a supplier before you qualify the part

Supplier qualification is the step that separates a part that works in isolation from a part that will work reliably across hundreds of units and multiple production batches. The standard sequence is to request samples against a specification, cross-check the supplier's own test data rather than accepting it at face value, run a small pilot batch, and apply Acceptable Quality Limit (AQL) sampling inspection on that batch to estimate the defect rate the supplier's process actually produces — not the defect rate their marketing material claims.

For the first production batch specifically, teams typically run a First Article Inspection (FAI) on a representative sample: a full dimensional check on the features that matter most — bearing fits, mounting concentricity, cable-routing clearances — a functional test that powers and commands every joint and sensor through its full range, and usually a burn-in run under nominal load while temperature, current draw, and position error are monitored. Because early builds carry more variability than a mature line, teams often combine FAI with heightened or 100% inspection on those early units rather than relying on the first-article sample alone. Units that fail are reworked or scrapped before they ever leave the factory. The resulting FAI report — photographs, measurement data, functional test logs — becomes the documented baseline that later batches are compared against, and it is often the single most valuable artifact a small manufacturing program produces.

Tooling and minimum-order-quantity economics

Robot volumes are small compared to consumer electronics, and that changes the economics of tooling. A custom injection-molded part might need a tool costing tens of thousands of dollars that only makes sense once volume is high enough to amortize it; at ten or a hundred units, machined parts or lower-cost tooling (aluminum rather than hardened steel molds, simpler single-cavity tooling) are often the more sensible choice even though the per-unit cost is higher. Component suppliers similarly set minimum order quantities that can be far larger than a small robot program actually needs, which forces a choice between over-ordering (tying up cash in inventory) or paying a premium for a below-MOQ order. Neither choice is free, and the right one depends on how confident the team is in near-term volume — a judgment call worth making deliberately rather than by default.

Quality systems and traceability

A production run introduces challenges that simply do not exist at prototype scale. Incoming material inspection verifies that the batteries, actuators, and sensors arriving from a supplier actually match the samples that were originally qualified — batch-to-batch variance from the same supplier is one of the most common sources of field failures. Production tooling — jigs for repeatable cable routing, torque-controlled fastening tools at every joint, automated test fixtures that run a full functional test in minutes per unit — is what makes a hundred units consistent instead of merely similar.

Traceability ties it together: each unit gets a serial number, and every sub-component — battery pack, actuator by joint position, compute board, sensor by type — is recorded against that serial number. When a failure surfaces in the field, a traceability system is what lets a team identify exactly which batch of components is implicated and issue a targeted recall, instead of guessing or recalling every unit ever shipped. Building this system before the first sale, not after the first field failure, is one of the clearest markers of a manufacturing program that is actually ready to scale.

The classic failure modes

A handful of mistakes recur across small-batch hardware programs regardless of what the product is. Single-sourcing a critical part — an actuator, a battery cell, a sensor — with no qualified second supplier leaves a program exposed to a single factory's capacity, quality, or business problems; qualifying a backup source, even one used rarely, is cheap insurance. Trusting a supplier's datasheet or verbal claim without independent test data is the second recurring mistake — the gap between a claimed specification and a verified one is exactly what supplier qualification exists to close. The third is leaving compliance or safety testing until late in the schedule: when a late test reveals a problem after tooling has already been committed and a production batch is on order, the fix is far more expensive than it would have been earlier, and the schedule slip lands at the worst possible time. None of these are exotic risks — they are the ordinary, well-documented ways small manufacturing programs go over budget and behind schedule, and each has a straightforward countermeasure applied early.