Researchers have developed a new method for safe and efficient navigation in crowded areas with heterogeneous shapes. The Shape-Aware Reinforcement Learned Model Predictive Control (SRL-MPC) approach uses reinforcement learning to adapt to changing environments and incorporates safety constraints based on geometric separation features. This allows the system to navigate complex scenarios without simplifying geometry or relying on handcrafted parameters, making it more suitable for dense crowd scenarios. --- Why it matters: This matters because current navigation systems often struggle with heterogeneous crowds and complex geometries, limiting their deployment in real-world scenarios. SRL-MPC's ability to adapt to changing environments and prioritize safety could improve the efficiency and reliability of autonomous robots and vehicles. Source: https://arxiv.org/abs/2608.21175