On 13 October, on the Startup Battlefield stage at TechCrunch Disrupt 2026, Industrial Next showed self-contained robotic stations that learn new tasks in simulation rather than from code. It also announced a collaboration with BMW Group Technology Office on dexterous manipulation.
Key takeaways
- The robotic stations are modular — reconfigurable, combinable and redeployable across product lines
- New tasks come from simulation and demonstration, not written scripts
- The company reports over 99.9% match accuracy and a cycle time under 3.3 seconds
- Stations already run in the production of one of the world's most popular smartphones
- The founders — Lukas Pankau and Allen Pan — previously worked at Tesla
A robot programmed for one thing
Classic industrial robots are programmed for a highly specific, repeatable task. Even a small change in the process or product breaks the program and costs downtime, pushing manufacturers to simplify products for repeatability. Retooling a line means reprogramming and physical tool changes — weeks or months.
The founders reached that conclusion at Tesla, where, as Pankau puts it, engineers designing cars were stumped in front of equipment from the 1980s.
We started this company because that is the kind of performance that we wish we had at Tesla, where we were bypassing robot stations because we couldn't program them effectively enough using traditional technologies.
Lukas Pankau, co-founder of Industrial Next.
Training in simulation instead of factory data
A flexible robot needs enormous training data, and robotics — unlike models trained on the internet — has no such datasets. The company's answer is a high-fidelity simulation engine close to real dynamics — an attempt at the classic sim-to-real problem. Inside it the robot gains a general understanding of the process and performs precision tasks "almost without seeing the parts at all".
The company's foundation models connect vision, task context and motion, with new skills arriving through teaching by demonstration, not a script.
The diagram orders the steps described by the company and TechCrunch. It does not present the models' architecture or the station's build.
| Dimension | Classic line | Industrial Next |
|---|---|---|
| A new task | reprogramming and physical tool changes | simulation plus demonstration |
| Retooling time | weeks or months | minutes |
| Product variant change | downtime risk, simplify the product | variants handled live |
What is already on the line
According to TechCrunch, some stations work on one of the most popular smartphones — threading a delicate flex cable?flex cable: A flexible printed circuit linking modules inside a phone. Easy to crease or snap, which is why assembly was long left to people. and snapping a tiny connector cap into place, tasks manufacturers had failed to automate for years.
The station stripping protective film off camera modules and buttons runs at or above human speed and, the company says, with 100% accuracy, even on a phone model and line it has never seen — at a cycle time?cycle time: How long a station takes per unit. It sets the pace of the whole line: the slowest station fixes the factory's throughput. under 3.3 seconds.
Why it matters
Manufacturers simplified products for repeatability because every variant cost weeks of engineering. Moving that cost into simulation changes not just retooling speed but what is worth building on a single line at all. The figures, though, come from the company and cover narrow assembly tasks, the customer stays undisclosed, and no third party has measured any of it — early deployments, not an industry standard.
What's next
- BMW Group Technology Office is evaluating the company's dexterous manipulation in manufacturing
- The stated goal is assembling a full car end to end with no human involvement
- The company does not disclose the name of the smartphone maker
Sources
- TechCrunch — Industrial Next wants factory robots to switch tasks in minutes, not weeks
- Industrial Next — AI-Defined Factories
