AI Robotics ·
Should AI recycling sorters be judged on both recovery and saleable purity?
Problem and evidence
TOMRA's November material describes adding camera-based deep learning to established sensors and monitoring separation quality [1]. Incheon reported that site-wide recovery rose from 73.7% to 75.2% after process improvements that included robots focused on clear PET [2]. These figures do not isolate the robots' contribution or establish the change in saleable purity.
| Dimension | International case | South Korea |
|---|---|---|
| Published operations | TOMRA describes classification and quality-monitoring capabilities | Incheon reports a 1.5 percentage-point gain in site-wide sorting |
| Interpretation | Accuracy needs retesting when input and line settings change | Concurrent changes prevent attributing the improvement to the robots alone |
Options and evaluation
A facility can procure on vendor accuracy specifications, or run paired trials and assess saleable weight alongside re-sorting costs. Log moisture and composition, match feed batches across existing and upgraded lines, and track downtime and rejected shipments. Reconfigure product classes or settings if additional recovered weight is offset by rework and lower sale prices.