Skild AI Unveils S1 Foundation Model to Give Robots a General-Purpose Brain
The 'omni-bodied' model allows robots to learn complex tasks from a single human video without additional training.
Skild AI has launched S1, a flagship robot foundation model designed to serve as a general-purpose brain for diverse robotic hardware. The system aims to eliminate the need for task-specific retraining by allowing robots to learn new behaviors through simple visual observation.
S1 utilizes in-context learning, which enables a robot to perform complex, long-horizon tasks—some lasting up to 10 minutes—by watching a single video of a human performing the action. This approach removes the requirement for traditional post-training or fine-tuning for every new application. To achieve this versatility, Skild AI trained the model on a diverse dataset including physics simulations, internet videos of humans, and teleoperation logs. CEO and co-founder Deepak Pathak explained that the company utilizes multiple data sources because the advantages of one compensate for the downsides of another.
A Hardware-Agnostic Approach
Traditionally, robotic AI has been limited by the need for massive, robot-specific datasets, meaning a model trained for one machine rarely worked on another. S1 is designed to be "omni-bodied," meaning it is hardware-agnostic and compatible across various forms, including humanoids, quadrupeds, and static robotic arms.
This shift toward a foundation model is an attempt to replicate the "ChatGPT moment" for robotics. Rather than rewriting the network's weights for a new task, users provide instructions—in this case, a video—within the model's prompt or context. Pathak noted that by adding a video of a human to the prompt, the robot can simply follow the observed action.
Scaling for Industry
To accelerate the deployment of this technology in real-world environments, Skild AI recently acquired assets from Fetch Robotics via Zebra Technologies. This move is intended to speed up the scaling of S1 within warehouse and industrial settings.
The financial backing for this ambition is significant. Since its 2023 founding, Skild AI has raised approximately $1.7 billion. This includes a $300 million Series A that valued the company at $1.5 billion and a subsequent $1.4 billion Series C round led by SoftBank.
The Path to Generalization
If S1 successfully generalizes across different physical bodies and learns complex tasks from simple visual prompts, it removes one of the primary bottlenecks in the field: the reliance on exhaustive, manual data collection for every new use case. Such a breakthrough could drastically accelerate the integration of general-purpose robots into both industrial workflows and domestic environments.
Industry observers will now be watching to see how S1 performs in uncontrolled, real-world settings beyond the initial demonstrations, and whether the model can maintain its efficiency as the complexity of the requested tasks increases.