Cosmos Policy shows how NVIDIA is using pretrained video models to simplify robot control and improve long-horizon task planning.
NVIDIA has introduced Cosmos Policy , a new approach to robot control that builds on its broader push toward world foundation models for physical AI systems.The framework is designed to simplify how robots decide what actions to take by adapting large video prediction models for control and planning tasks.
Policy rethinkIn robotics, a policy is the decision-making layer that translates observations, such as camera images and sensor data, into physical actions like joint movements or grasping motions. Traditional robot policies are often built as task-specific neural networks that require separate perception, planning, and control modules.These systems typically need large amounts of labeled data and custom tuning for each robot or environment.Cosmos Policy takes a different approach. Instead of designing a new control model from scratch, NVIDIA post-trains a pretrained video world model, known as Cosmos Predict, on robot demonstration data.The model already understands how the physical world evolves over time, having learned from large-scale video data.During post-training, robot actions, physical states, and task outcomes are treated as part of the model’s internal temporal representation, allowing it to predict both what the robot should do next and what will happen as a result.This design allows Cosmos Policy to jointly predict actions, future states, and expected task success within a single architecture.By relying on a single post-training stage, the framework reduces architectural complexity and avoids the need to stitch together multiple specialized models for perception and control.Benchmark resultsBenchmark results suggest the approach is effective. On standard robotic manipulation benchmarks, Cosmos Policy achieved high success rates across multi-step tasks that require long-term reasoning.In some cases, it matched or exceeded the performance of existing methods while using significantly fewer training demonstrations.This data efficiency is particularly important in robotics, where collecting real-world training data is costly and time-consuming. By leveraging knowledge already embedded in large video models, Cosmos Policy reduces the amount of robot-specific data needed to learn reliable control behaviors.Planning abilityAnother key feature of Cosmos Policy is its ability to perform planning at inference time. Rather than producing only the next immediate action, the model can generate and evaluate multiple candidate action sequences.By predicting the future outcomes and expected rewards of these sequences, the robot can select actions that are more likely to succeed over longer horizons.This planning capability allows robots to behave less reactively and more strategically when faced with complex tasks. The framework has also been evaluated on real-world robotic systems.In physical experiments involving bimanual manipulation, the policy was able to complete long-horizon tasks directly from visual input, suggesting that the approach can transfer beyond simulation into real environments.Cosmos Policy is part of NVIDIA’s larger Cosmos ecosystem, which focuses on building general-purpose world models for robots and autonomous systems. The broader goal is to provide shared infrastructure that helps machines understand, predict, and act within the physical world, reducing the need for task-specific engineering.Rather than acting as a rulebook or safety standard, Cosmos Policy serves as a technical foundation for robot control. Safety, compliance, and governance remain responsibilities of higher-level systems and regulators.As robots become more capable and autonomous, frameworks like Cosmos Policy highlight a growing industry effort to standardize the core decision-making layer that connects AI reasoning to physical action.
Cosmos Policy Nvidia Physical AI Robot Control Robotics World Models
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Rubio Defends Trump's Venezuela Policy, Addresses Foreign Policy ConcernsSecretary of State Marco Rubio defended President Trump's military operation in Venezuela, while also addressing concerns about U.S. foreign policy, including relations with Greenland, NATO, Iran, and China, during a Senate Foreign Relations Committee hearing.
Read more »
Tesla discontinuing Model S and Model X to make room for robotsTesla is ending production of the Model S and Model X to make room for robot production later this year.
Read more »
The Tesla Model S And X Are Dying Next Quarter To Make Room For RobotsIt's the end of the road for Tesla's long-running electric sedan and SUV. The automaker plans to make Optimus robots in their Fremont factory space.
Read more »
Tesla's Fastest-Growing Business Isn't Cars Anymore—Or Even RobotsThe company's car business may be struggling, but its energy storage division shows no signs of slowing down.
Read more »
Tesla ends production of Model S and Model X vehicles, will focus on robots in 2026Tesla pivots from electric vehicles to autonomous technology as Elon Musk announces the discontinuation of Model S and Model X to focus on mass-producing Optimus robots.
Read more »
Tesla is killing off its Model S and X cars to make robotsFind the latest technology news and expert tech product reviews. Learn about the latest gadgets and consumer tech products for entertainment, gaming, lifestyle and more.
Read more »