Skild AI가 단일 영상 시연으로 새로운 산업용 작업을 학습할 수 있는 로봇 기초 모델 S1을 출시했습니다.
Skild AI has released the S1 robot foundation model, designed to learn previously unseen long-horizon tasks from a single video demonstration. The model uses video as input to understand and execute tasks without updating its weights or undergoing task-specific retraining, a technique called in-context learning. Skild built S1 and conducted the research on NVIDIA AI infrastructure as part of a broader collaboration spanning synthetic data generation, model training, simulation and real-world physical AI deployment.
The robot addresses a fundamental challenge in industrial automation: manufacturing floors, warehouses and production lines constantly change, with shifting tasks, layouts and new products arriving, yet most robots cannot adapt without significant reprogramming. Deepak Pathak, cofounder and CEO of Skild AI, said that learning by experience rather than preprogramming represents the step change in robotics, noting that NVIDIA Isaac Lab and NVIDIA Cosmos technologies help Skild create the scalable, diverse experience its robots need to learn across many scenarios and embodiments.
The launch comes as Skild reached a $100 million annual revenue run rate just 10 months after its first commercial deployment. During that period, the company built more than 60 deployment partnerships with applications spanning manufacturing, logistics, inspection, security, food preparation and other domains.
Unlike traditional industrial robots built for fixed jobs, S1 operates differently. An operator records a video of the desired task and provides it to the model as a prompt. The model interprets the demonstrated intent, objects and sequence, then maps them into actions for the robot without requiring retraining, often for tasks not covered by its pretraining dataset. S1 can perform unfamiliar tasks lasting up to 10 minutes, including plant potting, pancake making, pour-over coffee brewing and kit assembly. These tasks can span dozens of manipulation steps and require the robot to compose skills in sequences it has not previously performed.
In testing, one plant-potting demonstration took just 11 minutes to go from video recording to autonomous execution on hardware. The model can adjust when objects move, recover from errors and combine skills in new sequences. In Skild's tests on new multistep tasks, S1 succeeded about 66 percent of the time at each step, compared with 9 percent for a similar AI system, more than a sevenfold improvement. Skild estimates that showing the robot one short video example can be as useful as giving it roughly 380 hands-on training examples, work that could take a person 50 to 100 hours to collect manually.
Skild, NVIDIA and Foxconn are now deploying the Skild Brain on dual-arm manipulators for high-precision assembly of NVIDIA Blackwell systems. In one demonstrated workflow, a robot installs a busbar and limit block, fastens 16 screws and adapts to disturbances across a multistep task, requiring precise motion, contact-aware control, sequence tracking and recovery when the scene differs from the plan.
NVIDIA accelerated computing powers Skild's ability to train its shared robot brain using simulation, human video, teleoperation and deployment data. NVIDIA Cosmos open world foundation models help diversify training data and turn video into structured descriptions, while NVIDIA Omniverse libraries and Isaac Sim provide physically based virtual environments for generating data, testing edge cases and validating behaviors. Skild uses Isaac Lab, an open modular robot learning framework powered by the Newton physics engine, to strengthen its brain through reinforcement learning, helping engineers accurately model physical parameters such as forces, contact, collision and pressure to reduce the simulation-to-reality gap. Skild and NVIDIA are jointly developing new GPU-accelerated simulation solvers to quickly and accurately model how robots physically touch, grip and manipulate solid objects, with these tools to be made available to all developers as part of Newton. NVIDIA Nsight tools help engineers find performance bottlenecks during training, and NVIDIA TensorRT optimizes inference so robots can respond quickly in the physical world.