The first sign is rarely a light. It is a 0603 landing a millimetre behind its pads on the first board of a batch, and the same millimetre on every board after it. By the time anything goes red you have scrapped a few boards.

On a Vertex 4 I have run short feeds, double feeds and a feeder that dropped off the RS-485 bus mid-job, plus one that was fine when I pulled it out and put it back on the shelf. The order below costs a pen and a pair of calipers before it costs a spare feeder.

The first sign was a 0402 that came back from the bottom camera sitting at an angle.

I was halfway through a 20-board batch of a small sensor board: nine 0402 decoupling caps, a dozen 0603s, two SOIC-8s. The machine is a Vertex 4, running OpenPnP. It ships with a 503 and a 505, and the 503 went into N1 for the small end of the BOM, which is the reasonable reading of those two nozzles if you have not thought about it much. I had not thought about it much.

If you've been sourcing PCB assemblies in 2026, you've felt the squeeze. Lead times that used to be 4-6 weeks are now stretching to 16-26 weeks. Assembly house pricing has climbed 20-40% since last year, and minimum order quantities are creeping upward as shops prioritize larger clients.

The root cause is familiar: component shortages continue to ripple through the supply chain, and the assembly houses that survived the 2024-2025 crunch are now capacity-constrained. They're cherry-picking jobs, and small-batch prototyping runs are the first to get deprioritized.

For hardware startups and small teams, this creates a strategic question worth answering: is it time to bring SMT assembly in-house?

The bottom camera is arguably the most important vision system on your desktop pick-and-place. It's what verifies that the component is correctly oriented on the nozzle before placement. Get the exposure wrong, and you're either blind to misaligned parts or drowning in false detections.

Most OpenPnP guides tell you to "adjust exposure until it looks right." That's not helpful. Here's what's actually happening and how to calibrate it properly.

Photon feeders use UUIDs (Universally Unique Identifiers) to communicate with OpenPnP. This is generally reliable, but it creates a specific class of problems when feeders share IDs, get reassigned, or conflict with each other. If you've ever seen OpenPnP detect the wrong feeder or fail to recognize a feeder that's physically connected, UUID conflicts are the likely culprit.

Here's what's happening under the hood and how to fix it.

Your nozzle has no vacuum and you can't pick anything up. Before you start unwiring the solenoid, check the actuator command — on the PikkoBot Vertex 4 that's the cause far more often than a hardware fault, and it's a one-field change.

Every hardware startup faces the same question: do you buy a desktop pick-and-place or invest in an industrial machine? The answer depends on where you are now and where you're heading. Here's an honest comparison with real numbers, not marketing copy.

"What component sizes can a desktop pick-and-place actually handle?" is the question we get most often, and the honest answer is that package size is usually not the limit. The limit is tape width, nozzle choice, and how much time you're willing to spend on the small stuff.

Here's the reference chart we use, and then the four things that actually decide whether your board runs reliably.

Fiducial detection is the backbone of accurate placement. When OpenPnP can't find your board's fiducials, the entire assembly process stops. Before you blame the software, work through these five fixes in order—they resolve the vast majority of detection failures.

Welcome to the PikkoBot blog. This is where we share what we're learning about desktop pick-and-place assembly, OpenPnP, and the practical realities of building hardware in-house.