OpenWAM: An Open Framework for Composable World-Action Models

OpenWAM is an open framework for world-action models that enables robots to predict future states and plan actions through a shared video foundation with configurable video-action interactions. The framework achieves significant performance improvements on robot control tasks, with success rates jumping from 68.4% to 97.8% on LIBERO-Long benchmarks, and demonstrates that counterfactual data improves dynamics models beyond demonstrations alone. By providing a common testbed with interchangeable components, OpenWAM allows researchers to systematically compare different designs for coupling prediction and control in robotic systems.