The field of embodied AI is witnessing significant developments in the area of world models, which are internal simulators that capture environment dynamics to support perception, prediction, and decision making. Recent research has focused on creating more realistic and interactive virtual environments, advancing the capabilities of intelligent vehicles, and improving the performance of robotic agents in complex scenarios. A key direction in this field is the development of unified frameworks that can seamlessly integrate diverse simulation paradigms, allowing for more efficient and effective training of agents. Another important area of research is the creation of large-scale datasets and benchmarks for evaluating the performance of world models in various tasks, such as pedestrian intent detection, off-road autonomous driving, and rough GPS-guided path planning. Noteworthy papers in this area include:

• Generalized Dynamics Generation towards Scannable Physical World Model, which presents a framework for integrating rigid body, articulated body, and soft body dynamics into a unified system.
• Zero-shot World Models via Search in Memory, which leverages similarity search and stochastic representations to approximate a world model without a training procedure.
• ProTerrain: Probabilistic Physics-Informed Rough Terrain World Modeling, which introduces a probabilistic framework for modeling spatially correlated aleatoric uncertainty over terrain parameters.