FlowMo-WM: A World Model with Object Momentum and Hidden Ambient Drift
Researchers have introduced FlowMo-WM, a novel visual world model designed for robot learning that accounts for object momentum and hidden ambient drift. Unlike previous models that focus on immediate control, FlowMo-WM can predict future states in environments with inertia and external forces like currents or wind. It achieves this by separating short-term object motion from long-term environmental influences, demonstrating improved accuracy in simulated aquatic vehicle scenarios. AI
IMPACT Introduces a more robust world model for robotics that can handle complex real-world dynamics like inertia and environmental drift.