Industries / Technology / IT hardware recycling
Dismantling
Dismantling is the work of taking old computers, phones, screens and servers apart, by hand or by machine, removing the batteries, leaded glass and other hazardous parts, and sorting what is left for reuse, the shredder or the smelter.
5
Milestones achieved, from 2009 to 2021.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2009
Shredders lose the gold
Researchers ran a substance flow analysis, a count of where each metal in the input ends up, at a full-scale plant that shreds and sorts electronic scrap before it is sent to smelters. Only 11.5% by weight of the silver, and 25.6% of the gold and of the palladium, reached output fractions from which they could be recovered, against 60% of the copper. The authors wrote that such plants are built to recover the metals present in bulk, such as iron and copper, and that trace metals are often lost.
LessonA plant tuned to recover metals by weight can lose most of the metals worth the most per gram.
Credit
Perrine Chancerel, Christina Meskers, Christian Hagelüken and Vera Susanne Rotter, in the Journal of Industrial Ecology
2016
A robot for one phone
In March 2016 Apple showed Liam, a line of 29 robots about 100 feet long, custom built for the iPhone 6. It took a phone apart in 11 seconds and could handle 1.2 million phones a year, by Apple's figures; an earlier version piloted from 2013 had needed 12 minutes a phone. Liam stopped work in May 2016.
LessonSpeed on one model was not enough, and Apple's next robot was built to handle many models.
Credit
Apple; described by Brian Heater of TechCrunch in 2024
2016
Leaded glass left behind
Closed Loop Refining and Recovery took in the cathode-ray tubes (CRTs) of old TVs and computer monitors, whose glass contains lead, and pledged to build a furnace to recycle the glass. It shut in 2016 and left material in Ohio and Arizona. Its landlord's 2017 federal suit counted 128 million pounds in two Columbus, Ohio, warehouses, 113 million pounds of it crushed CRT glass. In August 2025 a Franklin County judge ordered the company to pay $3 million in civil penalties; the Ohio Attorney General's office put the stockpile at 150 million pounds and said the building owners had spent $17 million over three and a half years to clean the sites.
LessonA recycler paid to take a material it cannot yet process can leave the bill to others.
Credit
Closed Loop Refining and Recovery; Garrison Southfield Park, the landlord; the Ohio Attorney General's office
2018
Robots punch phones apart
Apple's Daisy, shown in April 2018, could take apart nine iPhone models at up to 200 phones an hour, by Apple's figures. When TechCrunch visited the unit in Austin, Texas, in 2024, it handled 29 models. It chills each phone to −80 °C so the battery adhesive fails, then knocks the battery and the other parts out instead of unscrewing them, in 18 seconds a phone. Apple has two units, in Austin and in Breda, the Netherlands, and had offered other companies free licences to Daisy's patents; none had taken up the offer, TechCrunch reported.
Credit
Apple; reported by Brian Heater, TechCrunch
2021
Battery fires counted
In July 2021 the US Environmental Protection Agency (EPA) published a count, gathered mostly from news reports, of 245 fires caused or likely caused by lithium batteries at 64 waste facilities, electronics recyclers among them, in 28 states from 2013 to 2020, and said many more were likely missed. It found that batteries glued into phones, laptops and headphones slow dismantling sharply: an iPad with a glued-in battery may take 40 minutes to take apart and be worth at most $1 in scrap, a figure it attributes to Fowler (2018).
LessonA battery glued in to make a device thin becomes a fire risk and a labour cost at the recycler.
Credit
US EPA
Watch for
Batteries found by X-ray
The EU-funded GRINNER project, which ran from September 2022 to August 2025, says it built a prototype that passes waste electronics on a conveyor through energy-resolving X-ray detectors, and that its AI model, trained by a University of Essex team on synthetic and real images, reached "99 percent efficiency in battery detection". Its partners say they are developing a commercial system.
Achieved whenan outside party measures how many batteries the system misses in a recycler's normal feed, or a recycler reports a year of routine use with its miss counts.
Claimed by
the GRINNER consortium, in its final report to the European Commission covering March 2024 to August 2025
Open
01
Batteries in mixed loads
An open set of X-ray images of small electronics as they arrive at a recycler, several devices to a tray and overlapping, with every battery marked and checked against what the recycler found when it took each item apart. Then a scored test of battery detectors on it.
The briefCredit · unclaimed02
Robot opens laptop bases
A low-cost, open robot cell that removes the screws from the bottom cover of common laptops and lifts the cover off, trained from teleoperated demonstrations (a person steering the arm by remote control while it records), with the training recordings and the trained policy published.
The briefCredit · unclaimed
Open 01 · the brief
Batteries in mixed loads
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open set of X-ray images of small electronics as they arrive at a recycler, several devices to a tray and overlapping, with every battery marked and checked against what the recycler found when it took each item apart. Then a scored test of battery detectors on it.
Why it is still open
The one open X-ray set found, XBAT+ (December 2025, CC BY 4.0), built from devices gathered at an Irish recycler and at collection events in Limerick, holds 3,045 static images of 421 devices in 15 categories, labelled by device type and by whether a battery is present. GRINNER's 99% is its own figure, and its data is not public. Fraunhofer's DangerSort product page gives a throughput of up to about 30 metric tons an hour but no detection rates. A search of Hugging Face and GitHub in November 2026 found no X-ray set of mixed, overlapping e-waste loads.
Built with what exists
XBAT+ as training data; YOLO26 and RF-DETR, the detectors the XBAT+ authors scored, as baselines; Label Studio for marking batteries; the vision model in Liquid AI's LFM3 family, fine-tuned on one graphics card (see the Note "One GPU to fine-tune", a draft); an X-ray cabinet or inline scanner run by its licensed owner, a recycler or university. The team never handles damaged batteries, and no device is powered on or read: the images show hardware only.
First step
Download XBAT+, train a YOLO baseline on its X-ray images, and ask one recycler or university with an X-ray cabinet to image 50 trays of mixed small devices from one collection cage.
Done looks like
At least 2,000 X-ray images of mixed trays from at least two recyclers, each tray holding at least five devices, with every battery marked and confirmed by the recycler's dismantling record. Published for at least two detectors, one trained only on XBAT+: batteries missed per 1,000 battery-holding devices and false alarms per 1,000 devices, on a held-out set.
Open 02 · the brief
Robot opens laptop bases
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A low-cost, open robot cell that removes the screws from the bottom cover of common laptops and lifts the cover off, trained from teleoperated demonstrations (a person steering the arm by remote control while it records), with the training recordings and the trained policy published.
Why it is still open
Molg, a US start-up backed by ABB, says its robotic microfactories take apart laptops and servers on their own, but its announcements give no success rates. Fraunhofer IFF's iDEAR demonstrator removed a motherboard from a computer in early 2025. Learned policies have been tried on desktops: Chang Liu, Sibo Tian, Xiao Liang and colleagues fine-tuned OpenVLA models to remove memory and processors in December 2025 and found that they failed some critical steps, and in October 2026 their PACER method reached 87 successful approach-and-handoff trials out of 100 on motherboard parts. The Danish Technological Institute published 342 images for detecting laptop screen screws in February 2026, for perception only. No published learned policy or filmed tally for laptop cover removal was found.
Built with what exists
SO-101 leader and follower arms with Hugging Face's LeRobot software (see the Note "No download", a draft); a small electric screwdriver as the arm's tool; LeRobot's ACT policy as a baseline; NVIDIA's open GR00T 2 model, to test whether a few demonstrations per laptop model are enough (see the Note "Five demonstrations per task"); the Danish Technological Institute's images for screw detection. Laptops come from a licensed recycler with drives and batteries already removed, so there is no data and no live cell on the bench.
First step
Get five laptops of one model with drives and batteries removed, record 50 teleoperated screw removals with the SO-101 leader arm, and train a first ACT policy.
Done looks like
20 filmed attempts on each of five laptop models, one of them left out of training, with screws removed, covers lifted, time per laptop and any damage published in a tally beside the videos.