3D bin picking is the process of using a 3D vision system to locate randomly-oriented parts inside a container, then guiding a robot to pick them one at a time and place them into a machine, fixture, or conveyor. The "machine loading" part usually refers to loading CNC lathes, machining centers, grinders, presses, or inspection stations.
Three things have changed in the last three years that make bin picking viable for medium-size factories:
Result: systems that used to require a $300K budget and 6-month integration now ship in 6–10 weeks at $80K–150K.
The single biggest mistake factories make is over-specifying the camera. A 5-megapixel camera doesn't pick parts better than a 1.3-megapixel one — it just costs more and runs slower.
What actually matters:
For metal parts 20–300 mm, a mid-range structured-light camera (like the ones we use at AIMA ROBOT) hits the sweet spot: 0.3 mm depth accuracy, 1.2 s cycle, works under normal factory lighting.
Most factories buy robots that are too fast and too small. A 7 kg payload cobot sounds great until you realize the gripper weighs 2.5 kg and the part 3 kg — leaving 1.5 kg of margin, which the robot controller will complain about every cycle.
Sizing rules of thumb:
| Part weight | Recommended robot payload | Typical reach |
|---|---|---|
| < 0.5 kg | 5 kg cobot | 800–1000 mm |
| 0.5–3 kg | 10–16 kg cobot or 6-axis industrial | 1200–1500 mm |
| 3–15 kg | 20 kg industrial 6-axis | 1500–2000 mm |
| > 15 kg | 35+ kg industrial 6-axis | 2000+ mm |
Also remember: the robot needs to reach both the bin and the machine chuck. Measure the actual distance in your cell, not the catalog number.
We've seen more bin picking projects fail at the gripper than at any other component. The reason: the gripper must handle the full range of part orientations the camera detects.
The three questions to ask your gripper supplier:
For cylindrical metal parts, magnetic grippers with adjustable pole pieces work well. For prismatic parts, 3-jaw pneumatic grippers with custom-machined fingers. For delicate parts, vacuum grippers with multiple suction cups and individual vacuum sensing.
Hardware is only half the system. The software stack has three layers, and you need all three working reliably:
| Layer | What it does | What to look for |
|---|---|---|
| Vision processing | Converts 3D point cloud into pickable part poses | Handles shiny/rusty/oily surfaces; processing time < 500 ms |
| Path planning | Computes collision-free robot trajectory from bin to machine | Real-time re-planning when part shifts; singularity avoidance |
| Cell control | Coordinates robot, machine door, chuck, conveyor | Standard protocols (EtherCAT, PROFINET, Modbus); error recovery |
Ask vendors for a live demo with YOUR parts, not their demo parts. If they can't show you a working pick with your actual parts in their lab within 2 weeks, walk away.
Here are actual numbers from systems we've deployed (not marketing numbers):
| Application | Part | Cycle time | First-pick success rate | Uptime (3 months) |
|---|---|---|---|---|
| CNC lathe loading | Steel shafts, 80–300 mm | 14 s | 97.2% | 99.1% |
| Machining center tending | Aluminum housings, 150×120×80 mm | 22 s | 95.8% | 98.7% |
| Grinder loading | Hardened steel rings, 40–120 mm | 18 s | 96.4% | 98.9% |
Key takeaways:
Most vendors show you a spreadsheet with fake numbers. Here's the formula we use with customers:
Annual savings = (labor cost per shift × shifts replaced) + (machine uptime gain × value per hour) − (system operating cost)
Worked example — CNC lathe cell, 2 shifts:
Annual savings = ($18,000 × 1.5) + (2,080 hrs × 12% × $85) − $4,200 = $27,000 + $21,216 − $4,200 = $44,016/year
System cost: $98,000
Payback period: 26 months
If your payback calculation shows less than 18 months, double-check the assumptions — it's probably too optimistic. If it shows more than 36 months, the application may not be right for automation yet.
Engineers always want to automate the most difficult part first — the one that's tangled, oily, or has 15 variants. Don't. Start with your simplest, most consistent part. Get the cell running reliably, then expand. We've seen projects fail because the team spent 6 months trying to solve the hardest part and never got to the easy ones that would have delivered ROI.
The vision system can only pick what it can see. If parts are stacked 5 layers deep and tangled, no amount of AI will help. Sometimes the answer is a simple vibratory pre-feeder or a redesigned bin with internal dividers — not a more expensive camera.
The robot needs to talk to the CNC machine: open the door, clamp the chuck, start the cycle, detect completion. This requires proper PLC integration, not just hard-wired I/O. Budget 15–20% of the project cost for controls engineering. Skipping this is the #1 reason cells sit idle after the vendor leaves.
We've deployed 3D bin picking systems across automotive, hardware, and precision machining industries. Our process:
We don't take projects where we don't believe we can hit 95%+ first-pick success rate. About 30% of inquiries we receive are redirected to simpler, cheaper solutions — because a failed bin picking project helps no one.
Have a specific application in mind? Send us photos of your parts and bin. We'll tell you within 48 hours whether bin picking makes sense for your case — and if not, what to do instead.
→ Request Free Feasibility Assessment