A Robot That Learns From Watching, Not From Months Of Training
Skild AI's S1 foundation model, detailed by NVIDIA on September 10, learns a new robot task from a single video demonstration rather than the months of task-specific data collection and retraining that industrial robotics has relied on until now. An operator records a video of the task once; S1 uses that recording as its prompt, works out the objects involved, the order of steps and the intent behind them, and maps the sequence onto robot actions without updating its underlying weights.
NVIDIA and Skild say one video demonstration is worth roughly 380 hand-collected training episodes, and in a plant-repotting test the whole path from materials arriving to the robot working autonomously took 11 minutes.
The First Job It Learned Was Assembling NVIDIA's Own Chips
Skild, NVIDIA and Foxconn have deployed the Skild Brain on dual-arm manipulators for precision assembly of NVIDIA's own Blackwell systems, with one demonstrated workflow showing a robot installing a busbar and limit block, fastening 16 screws, and adapting when the scene is disturbed. That is exactly the kind of long, multi-part precision work that contract electronics manufacturers currently pay skilled line workers to do by hand.
| Metric | S1, in-context learning | Comparable system |
|---|---|---|
| Success per step, unseen long tasks | 66 percent | 9 percent |
| Success on tasks seen in training | 96 percent | Lower, language-conditioned baseline |
| Training data needed per new task | 1 video, about 380 demo-equivalents | 50 to 100 hours of teleoperation |
Skild has reached 100 million dollars in annual revenue run rate just 10 months after its first commercial deployment, with more than 60 partners spanning manufacturing, logistics, inspection, security and food preparation. Skild's cofounder and chief executive Deepak Pathak has said learning from experience rather than preprogramming is the real shift now happening in robotics.
The Assembly-Cost Argument For Moving Production Just Got Weaker
Contract electronics assembly has clustered in low-wage regions for decades because precision hand-work is expensive to relocate: a new site needs a trained workforce, and training a workforce takes time. A robot that can pick up a comparable assembly task from one recorded demonstration and be working autonomously within minutes removes exactly that constraint.
If S1's numbers hold up outside Skild's own demonstrations, the deciding factor in where precision electronics assembly happens shifts away from local labor cost and toward robot-hours, power availability and how close a facility sits to the chips and components it assembles - variables a European or UK manufacturer competing for the same contracts can actually control.
What This Means For A European Contract Manufacturer
A European or UK electronics contract manufacturer should treat this as an early signal rather than an immediate threat: Skild's own numbers are self-reported and the Foxconn deployment is one demonstrated workflow, not a factory-wide rollout.
But the direction is worth planning around. If in-context robot learning keeps closing the gap NVIDIA and Skild describe, competing for the next generation of precision-assembly contracts will depend less on wage costs and more on who can integrate video-taught robots fastest, and on sitting close enough to a customer's chips and power supply to matter. European manufacturers with strong automation engineering and reliable industrial power have a real opening, provided they start building the integration skills now.
Servola Journal
We do this for everyone trying to keep up with what technology is doing to our lives. The people who build it, and the people it happens to. The Servola Journal exists so that what we learn belongs to all of them.
Nobody pays us for this. No ads, no paywall, free to everyone. We just believe that understanding what's happening to all of us shouldn't depend on who can afford to pay for it.
If it gave you something today, tell us to keep going. Follow us, leave a like, or write a positive comment. We read every one, and they are what keeps us going.
Read next: An Open Robot Brain Just Shipped, and Data Is the Moat | Bosch Sold Its Robot to the Startup It Funded
