Skild AI lanza el modelo de robot S1 entrenado a partir de un solo vídeo

que paso

Skild AI launched S1, a robot foundation model designed to learn new tasks from a single video demonstration. The system uses NVIDIA infrastructure and is targeted at industrial environments where workflows change frequently. Skild reported that S1 achieved a 66% success rate on new multistep tasks in benchmarks, compared to 9% for a similar system, and stated the company has reached a $100 million annual revenue run rate.

Skild AI has launched S1, a robot foundation model that learns new tasks from a single video demonstration. The system was developed using NVIDIA AI infrastructure and is designed for industrial settings where workflows, layouts, and product lines change regularly. According to Skild, the model uses a video prompt to interpret a task and carry it out without changing its weights or undergoing task-specific post-training.

An operator can record a video of a job and provide it to the model as an example. The system then identifies the intended sequence, the objects involved, and the actions required for the robot on site. Skild stated that S1 can handle previously unseen tasks lasting up to 10 minutes, with examples including plant potting, pancake making, pour-over coffee brewing, and kit assembly. In one plant-potting test, Skild said it took 11 minutes to go from recording a demonstration to autonomous execution on hardware.

Skild published benchmark figures alongside the launch, stating that in tests on new multistep tasks, S1 succeeded about 66% of the time at each step, compared with 9% for a similar AI system. The company also estimated that one short video example can be as useful as roughly 380 hands-on training examples, which could take 50 to 100 hours to collect manually. Skild reported reaching a USD $100 million annual revenue run rate within 10 months of its first commercial deployment and building more than 60 deployment partnerships.

The launch includes a collaboration with NVIDIA and Foxconn, where the Skild Brain is being deployed on dual-arm manipulators for high-precision assembly of NVIDIA Blackwell systems. In one workflow, a robot installs a busbar and limit block, fastens 16 screws, and adapts to disturbances. NVIDIA is using its Cosmos models to diversify training data and convert video into structured descriptions, while Skild uses NVIDIA Omniverse libraries, Isaac Sim, and Isaac Lab to train and validate the model before deployment.

Detalles de la fuente: itbrief.co.nz ↗

Por qué es importante

This launch represents a significant shift in industrial robotics by reducing the data and time required to deploy robots for new tasks. By using video prompts instead of extensive retraining, Skild aims to lower the barrier for dynamic manufacturing and logistics environments. The reported revenue milestone and partnerships with NVIDIA and Foxconn suggest practical commercial viability for this approach, potentially accelerating the adoption of flexible automation in sectors that previously relied on fixed, preprogrammed routines.

The ability to learn from a single video demonstration addresses a major bottleneck in industrial robotics, which traditionally requires extensive data collection and retraining for new processes. This approach could significantly reduce the time and cost associated with deploying robots in dynamic environments such as manufacturing and logistics.

The reported 66% success rate on new multistep tasks, if independently verified, would represent a substantial improvement over existing systems. This metric suggests that S1 may be capable of handling complex, multistep tasks that require the robot to combine skills in a sequence it has not performed before, potentially expanding the range of automatable jobs.

The collaboration with NVIDIA and Foxconn indicates a practical application of this technology in high-precision assembly. The use of NVIDIA's software stack, including Cosmos, Omniverse, and Isaac Lab, highlights the integration of simulation and real-world deployment, which is crucial for ensuring reliability and safety in industrial settings.

Skild's reported $100 million annual revenue run rate and 60+ deployment partnerships suggest that this technology is not just a research prototype but has achieved commercial traction. This could signal a broader shift in the robotics industry toward more flexible, AI-driven solutions that can adapt to changing production needs without extensive reprogramming.

Qué ver a continuación

Monitor independent verification of the 66% success rate and the 11-minute deployment time, as these metrics are currently self-reported by Skild AI. Watch for broader availability of the NVIDIA Newton physics engine solvers mentioned in the collaboration, and observe whether the Foxconn deployment scales beyond the initial high-precision assembly tasks. Additionally, track how Skild handles data privacy and customer agreements regarding the use of deployment data for broader model training.

Independent verification of the 66% success rate and the 11-minute deployment time is crucial, as these metrics are currently self-reported by Skild AI. Third-party testing will be necessary to confirm the model's performance in real-world industrial settings.

The availability of the NVIDIA Newton physics engine solvers to developers could have broader implications for the robotics industry. These solvers, which model how robots touch, grip, and manipulate solid objects, may enhance the capabilities of other robotics companies and accelerate the development of more advanced robotic systems.

The scale and success of the Foxconn deployment will be a key indicator of the technology's practical viability. If the Skild Brain can reliably perform high-precision assembly tasks in a large-scale manufacturing environment, it could serve as a proof of concept for broader adoption in the electronics industry.

Data privacy and customer agreements regarding the use of deployment data for broader model training will be important considerations for potential customers. Skild's approach of using commercial deployment data to improve the model could raise concerns about data ownership and confidentiality, particularly in sensitive industrial applications.

Guías y cuestionarios relacionados