The robot was not the bottleneck

Bales were being rejected for exceeding the contamination limit their buyer accepts. The plant manager proposed AI-guided robotic sorters on the quality-control line, and had quotes.

A single-stream facility taking mixed household recycling from several collection rounds, selling baled fibre into an export market with a strict contamination specification.

Mechanical screens, near-infrared optical sorters on the main lines, and a manual quality-control picking station at the end.

• Contractual: the offtake specification is fixed; a bale is either in spec or rejected.
• Physical: the picking station has a fixed belt speed and a fixed footprint.
• Cost: capital is available this year and not next, which was itself driving urgency.

• fact

  Optical sorters run at roughly 600 picks per minute against about 60 for most single-arm robots. Current models reach 70 on mixed plastics and paper, and the fastest multi-arm systems are reported far higher, with one vendor claiming over 200 picks per minute and another 6,000 picks per hour across three arms. The trade press treats the two as complementary rather than as substitutes: robots earn their place at quality-control positions.

  Source: Recycling Today, ‘Sorting it all out’; Waste Today, optical sorters vs robots

• fact

  Facilities meeting strict export fibre specifications report contamination limits around 0.5%, a level driven by the buyer, not the machine.

  Source: Recycling Today, MRF optical sorting case reporting

• assumption

  Contamination was assumed to be created on the sorting line. It had never been sampled by collection round.

  No external source: stated for the example

• hypothesis

  If contamination varies by incoming round, the cheapest fix is upstream of the plant entirely.

  No external source: stated for the example

1. 01Where does contamination enter, at the kerbside, in the bunker, or on the line?
2. 02What is the composition by collection round? Nobody had sampled it.
3. 03If a robot adds 60 picks a minute to a line already doing over a thousand, what fraction of the problem does it address?
4. 04What does a rejected bale cost, including the return haul?

• Buy robotic sorters

  AI-guided picking arms on the quality-control line.

  What it costs: Large capital cost, a maintenance regime the plant has never run, and it addresses the end of the line rather than the source. Industry reporting puts robots at about a tenth of an optical sorter’s pick rate, so the case rests entirely on the picking line being the binding constraint.

• Sample and act upstream

  chosen

  Two weeks of bunker sampling by collection round, then targeted communication and contract terms for the worst rounds.

  What it costs: Depends on the collection authority, which the plant does not control, but costs almost nothing to find out.

• Slow the line

  Reduce belt speed at the quality-control station.

  What it costs: Immediate quality gain, direct throughput loss. Kept as the fallback.

Build

Robots: a capital line big enough to need board approval. Sampling: two weeks of an existing operative’s time.

Run

Robots add maintenance, spares and a vision system whose accuracy drifts with the material mix. That is a permanent cost against a variable input.

Change

If the offtake spec tightens again, a robot tuned to today’s contaminants is not obviously the answer to tomorrow’s.

Exit

High for the robots, which are installed equipment with a thin second-hand market. Effectively nil for sampling.

Do not buy robots this year. Sample the incoming material by collection round first, and keep line-speed reduction as the interim control.

Why

The proposal treated a capital purchase as the answer before the constraint had been located. The optical sorters already on site outpace a robotic picker by roughly ten to one, so the marginal picks a robot adds are small unless the picking line is the actual limit, and no evidence said it was.

Why not the alternatives

• Buy robotic sorters: Unproven that the picking line is the binding constraint, and the equipment already installed operates at a far higher pick rate.
• Slow the line: Retained as a control, but it trades throughput permanently for a problem that may originate upstream.

• A year of capital availability passes unused. That is accepted, because spending it on the wrong constraint is worse.
• Some throughput is given up in the interim.

A vision system on the quality-control line would have added a new maintenance discipline to a plant that does not have one. It did not earn its place on the evidence available.

A sampling protocol, a clipboard, and a fortnight. Results by round, shared with the collection authority.

The facility learns where contamination actually enters, which determines whether the fix is a machine, a contract, or a leaflet. Rejection rate is the measure either way.

No outcome is claimed. This is an illustrative example, so there is nothing measured to report, and a real engagement would state what happened and how it was verified.

• Find the constraint before you fund the solution. A faster machine downstream of the bottleneck buys nothing.
• Compare like with like, and quote the whole range: a single-arm robot near 60 picks per minute against 600 on the optical sorter already installed, then check whether the fast multi-arm systems change the answer before ruling them out.
• The cheapest experiment in this engagement cost two weeks and a clipboard, and it decided a capital programme.