Model-Based Control API supports online planning with offline learning

AI Impact Summary

New capability introduces a model-based control module that plans actions online while learning from offline data. By leveraging a learned dynamics model, it enables safe, sample-efficient exploration and reduces the need for extensive live experimentation. This will primarily affect teams building autonomous agents, robotics, or optimization workloads that require rapid iteration with restricted online data, and may necessitate changes to data pipelines and offline training workflows to supply representative scenarios.