World Action Models (WAMs) aim to combine a prediction of how the world changes with the generation of actions for robots. WAMs require both a prediction of the future state and the ability to generate actions based on that prediction. This concept is contrasted with Vision-Language-Action (VLA) models, which focus on perception and control. Researchers explore WAMs to determine whether robots should react directly to what they see or imagine future outcomes. Different WAM architectures, such as cascaded or joint models, are being investigated to advance this area.
Groceryshop 2026 highlighted that retail robots are ready to scale by utilizing artificial intelligence. Companies like Simbe Robotics are rolling out thousands of robots, focusing on making their data actionable through analytics. Badger Technologies is shifting its focus to retail intelligence, aiming to provide actionable insights on shelf conditions for decision-making. Other innovations focus on new deployment models, like ShelfOptix's approach for smaller retailers, and the next evolution towards end-to-end shelf automation. The future involves robots moving from data collection to actively performing tasks like product picking to improve store execution.