The article discusses a new method to predict future occupancy in busy city areas for safe self-driving cars. By using a special network, the system can forecast where other vehicles will be and how they will move, even in complex traffic situations. This approach can predict occupancy up to 3 seconds ahead, outperforming current methods. The network can understand the environment without needing detailed maps or explicit models of moving objects. The researchers have shared their dataset to help others study this topic further.