Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
S1 is Skild AI's flagship robotic foundation model, built as an in-context learner. Instead of specifying a task through language or collecting hours of data to fine-tune a specialist policy, a user shows the robot a video demonstration of the task, and the model executes it without fine-tuning or post-training.
According to Skild AI, S1 is pre-trained on episodic data where the task is specified only through an in-context demonstration, which may come from a different scene, viewpoint or embodiment. The model must therefore infer the demonstrator's intent, functional correspondences and task progress. The company reports in-context execution of tasks up to 10 minutes long that were not seen during pre-training. Creator coverage mentions demos such as repotting a plant, pour-over coffee, assembling a kit and flipping a pancake, using a two-armed robot and demonstrators wearing wrist cameras.
The announcement is the first in a series of posts; later posts are to cover training details. S1 is built on NVIDIA AI infrastructure. No public weights, API, pricing or documentation were found, so it appears to be a research and announcement-stage model rather than a generally available product.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Robotics foundation model that learns a new task from one video demonstration without extra training. Tested on repotting a plant, pour-over coffee, assembling a kit and flipping a pancake, with tasks up to 10 minutes long.
A robot foundation model that learns 10-minute tasks never seen before from a single video prompt, with no fine-tuning. Demonstrators wear wrist cameras and the two-armed robot reproduces the task.
Closest neighbours by meaning, not just by category.
Research · Skild AI
Skild AI research result: humanoid soccer policy trained by self-play in Isaac Sim on the S1 model, transferred to a real robot.
Model · Generalist AI
Robot foundation model from Generalist AI that learns new tasks in-context from a short demonstration.
Research
Research framework letting a frozen VLM agent learn robot tasks in context from videos, images and feedback, no fine-tuning.
Model · Perceptron
Perceptron's vision-language model for physical AI, covering perception, physical reasoning and robot action prediction.
Model · Xiaomi
Open-weights Vision-Language-Action robot foundation model from Xiaomi, pre-trained on 100K+ hours of UMI trajectories.
Model · Google DeepMind
Google DeepMind family of robot models: a VLA, ER 2 for reasoning and planning, and On-Device 2 for offline use.
