A battery teardown business that never lets the robot walk
A humanoid picked battery screws 87 percent of the time standing still, and 37 percent after walking to the job. Build the business that keeps the machine still and moves the pack instead.
A humanoid robot picked screws from a real Hyundai Ioniq 5 battery pack 86.7 percent of the time standing still, and 36.7 percent after walking into place. The lesson is that precision needs a fixed platform. Therefore sell battery teardown as a service, where the pack travels to a bolted-down station and the machine never moves.
The finding this rests on
A humanoid robot was asked to pick loosened screws out of a real electric car battery pack. The pack was a Hyundai Ioniq 5 unit, 400 kilograms, about 1.2 metres by 2 metres. The robot was tested three ways, with 30 tries each way.
Hanging in a hoist, it got the screw 96.7 percent of the time. Standing on its own feet, 86.7 percent. After walking itself into place, 36.7 percent. The full article is at walking to the job cut a robot’s screw grabs from 87% to 37%.
The hands were not the problem. The feet were. Over 20 walks of 6 metres, the robot stopped an average of 13.0 centimetres from its target. For a screw in a hole, 13 centimetres is a mile.
We score this paper 6 out of 10. That is one of the higher scores in our index. It is a real robot on a real pack, with staged conditions and reported confidence intervals, and the code is public. It is also a preprint with no journal named and no citations yet. Therefore trust the shape of the result. Treat the exact percentages as early.
Who has this problem
Picture a vehicle dismantling yard on the edge of a mid-size city. Nine staff. Two of them are trained to work on high voltage packs. Old electric cars arrive on a low loader most weeks.
A pack comes in with an unknown charge. It can catch fire. It can give off toxic fumes. It can shock the person holding the spanner. So the yard treats every pack as live.
Hundreds of screws come out of each pack. That is a whole morning for one trained pair, in gloves and a face shield, in a bay nobody else may enter. The yard would rather have those two people on gearboxes and wheels. Instead they are turning fasteners.
What you would sell them
Sell teardown by the pack, not a machine. The yard ships you the pack. You ship back sorted modules, loose hardware and a photo log of every step.
On Monday morning nobody at the yard is standing in a sealed bay with a torque gun. The pack is on your floor, in your frame, under your insurance.
The finding is why the shape of this business matters. A robot that walks to the job loses more than half its grabs. A robot that never moves keeps almost all of them. Therefore build the version where the pack travels and the machine stays bolted down.
What to charge
Test it at 400 dollars a pack for the fastener step alone. That is a hypothesis, not a price list. You are trying to learn whether a yard will pay anything at all.
Reason it out against what the yard already spends. Two trained people in a sealed bay for a morning is the thing you replace. You know what an hour of your own crew costs. Put your number in and compare. If 400 dollars sits under half a morning of that pair, the yard has a reason to listen.
Do not price against the robot. The robot in the paper needed an average of 75 seconds per screw, from spotting it to holding it. A person is much faster than that. Anyone pricing robots against wages today should start from that number.
How you would build the first version
Build no robot in month one. Rent one bay with a roller door. Buy a hoist, a rolling frame that clamps a pack flat, a torque gun, and two cameras on stands.
Hire one trained person. They do the teardown by hand. The cameras record every pack, screw by screw, with a timestamp.
That video is the real product of version one. It tells you how many fasteners each pack design has, where they hide, and which ones fight back. It is also the layout study for any arm you add later.
Do not build a walking robot. Do not build a custom hand. The team in the paper used an open-source gripper made from printed and off-the-shelf parts for about 460 dollars, and it senses grip force in steps of 0.08 newtons. Parts like that already exist. What does not exist is your proof that yards will send you packs.
The one-week test
- Day one. List every vehicle dismantler and scrap yard within a two hour drive. Mark the ones who mention electric cars at all. Aim for 30 names.
- Day two. Call them. Ask one question. What do you do today with a high voltage pack that comes in on a car? Write the answer down word for word.
- Day three. Sort the answers into three piles. They pay someone to take it away. They store it and hope. They strip it themselves. Only the third pile is your customer.
- Day four. Go back to the third pile with one page. It says this. You ship us the pack. We ship back sorted modules and hardware. 400 dollars, and we are testing that price. Ask for a signed order for one pack.
- Day five. Count signed orders. One paid order from a yard you had never met is a yes. Zero is a no, and you have spent a week, not a year.
What would kill this
Speed. Seventy-five seconds a screw is slow. If you add a robot too early, you lose to a person with a torque gun. The yard will see that inside a week.
Pack variety. Packs come in many designs, so a rig built for one pack does not suit the next. The paper makes that point about fixed machines. Your clamping frame is a fixed machine. Every new design is a new setup.
Moving the hazard. A pack with an unknown charge is just as dangerous on your floor. Getting it to you is its own problem. Storage, fire rules and insurance may cost more than the job earns.
The research may not hold. This is one robot, in one lab, on one model of pack, with 30 tries per condition. The robot never unscrewed anything. A separate gantry machine loosened every screw first. One grasp method scored 100 percent in simulation and 0 percent on the real pack, so simulation did not predict real life. Two of the authors declare a stake in a company working on humanoids for manufacturing. If you build the automated version of this business, you are betting that the fixed platform result holds in a working plant.
What would settle it is a whole pack taken apart end to end, in a real recycling plant, timed against a human crew, over weeks. Until someone publishes that, sell the service and let the robot wait. BOLT THE MACHINE DOWN BEFORE YOU TRUST IT.
Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.
- Grasp success on a real battery pack fell from 86.7 percent standing to 36.7 percent after the robot walked itself into place, so the platform matters more than the hand.
- A teardown service moves the pack to a fixed station, which removes the exact failure the paper measured.
- Start with one bay, one trained person and two cameras, because the video log tells you what any future machine has to do.
Questions people ask
why sell a service instead of selling the robot?
Because the paper shows the robot loses most of its grabs once it walks to the job. A service keeps the machine in one place and moves the pack instead. You also learn what each pack design needs before you spend money on automation.
is a humanoid robot needed for this at all?
Not in version one. The first version is a bay, a hoist, a clamping frame, a torque gun and two cameras, run by one trained person. The robot is a later decision, and only if the recorded times say it pays.
how strong is the evidence?
We score the paper 6 out of 10. It is a real robot on a real 400 kilogram pack with 30 tries per condition and public code. It is also a preprint with no journal named and no citations yet, so treat the exact percentages as early numbers.
what is the biggest risk?
Speed. The robot in the paper took an average of 75 seconds per screw, and a person is much faster. Add hazard rules for packs with an unknown charge, and the margin can vanish before the work starts.
Mark Alex
Mark Alex is the founder and Managing Director of Real Biz Digital, a technology company operating out of Nairobi since 2018. He works in agentic AI and the Model Context Protocol, AI governance, enterprise software architecture and cybersecurity. He holds an MSc in Mechatronical Engineering from Obuda University in Budapest and a BSc in IT, Forensic Technology and Cybercrime, from USIU-Africa in Nairobi, and has published IEEE conference research on an AI-powered digital twin for greenhouse systems. He is the author of seven books. Between 2020 and 2024 he mentored more than 200 university students and interns in Nairobi. He writes every Just Out Tech article from the original research paper.