S1
About
About S1
S1 is the named flagship in-context-learning model within Skild AI’s broader Skild Brain platform. A user provides one video demonstration, and S1 conditions its robot policy on that example to perform the task without updating model weights. Skild demonstrates the approach on seen and unseen manipulation tasks including plant potting, pancake preparation, pour-over coffee and kit assembly.
The company reports unseen tasks lasting up to 10 minutes and a sevenfold improvement over language-only prompting on its evaluation set. Its plant-potting example spans an 11-minute demonstration-to-execution workflow. Skild says S1 is already operating with commercial partners, while public model weights, code and training data have not been released.
Capabilities
Source-backed capabilities
- Object manipulation
Research demonstration - 2026-08-18 - S1 applies in-context learning to seen and unseen manipulation tasks, including plant potting, pancake preparation, pour-over coffee and kit assembly.
Source
Observed use
Exact-product configurations are shown without inferring supply, purchase, customer status, standard fit, deployment or partnership.
- G1
Research Experiment - Controlled By - Mixed - A Unitree G1 played a physical football match using a Skild S1-derived policy post-trained through simulated self-play.
Source
Research
- Physical Self-Play
2026-09-23 - Research Publication - A scoring-only self-play objective produced dribbling, shielding, tackling and recovery behaviours after 140 simulated years of play in NVIDIA Isaac Sim.
Source - Introducing S1: In-Context Learning for Robotics
2026-08-18 - Research Publication - S1 applies in-context learning to long-horizon robot manipulation, with demonstrations spanning plant potting, cooking, beverage preparation and assembly.
Source
Timeline
- Skild AI publishes physical self-play result with humanoid transfer
2026-09-23 - Research Publication - The S1-based policy was rewarded for scoring rather than for hand-designed dribbling or tackling behaviours. Dribbling, shielding, tackling and recovery emerged as the policy competed against successive versions of itself in NVIDIA Isaac Sim. Skild says it is extending the same post-training approach to everyday robot tasks beyond sport.
Source - Skild AI publishes S1 in-context robot learning model
2026-08-18 - Research Publication - S1 applies in-context learning to seen and unseen manipulation tasks, including plant potting, pancake preparation, pour-over coffee and kit assembly. Skild reports unseen tasks lasting up to 10 minutes and a sevenfold improvement over language-only prompting on its evaluation set. The official release demonstrates transfer from a human video example to robot execution, including an 11-minute demonstration-to-execution plant-potting workflow.
Source
Context
Comparable Systems
- Being-H0.8
BeingBeyond - Robotics Foundation Model
- Helix
Figure AI - Robotics Foundation Model
- FusionMax
Muks Robotics AI Pvt. Ltd. - Robotics Foundation Model
- GEN-1
Generalist - Software
- GENE-26.5
Genesis AI - Robotics Foundation Model
- Lumo 2
Astribot - Robotics Foundation Model
Related Systems
- DYNA-1
Dyna Robotics - Robotics Foundation Model