Waymo and Tesla are pursuing similar paths in developing self-driving technology, despite common misconceptions about their approaches. Both companies have moved away from traditional modular methods to embrace transformer-based, end-to-end architectures. Waymo’s recent developments emphasize this shift, as they now use a foundation model trained in an end-to-end manner, akin to Tesla’s system. Waymo’s model, called EMMA, is built on Google’s Gemini and maps raw camera data to driving outputs. However, while promising, EMMA is not yet ready for commercial deployment due to challenges like high computational demands and poor spatial reasoning. Waymo's current self-driving technology combines the EMMA-style vision-language model with a hybrid system that includes a sensor fusion encoder. This encoder enhances speed and accuracy by breaking down scenes into individual objects, leveraging lidar sensors for precise distance measurements. Unlike earlier systems that relied on human-defined object attributes, Waymo's model learns to represent objects through data-driven processes. This approach captures essential information relevant to driving, which can be more nuanced than traditional coding methods. The article highlights how both Waymo and Tesla are refining their systems to handle the complexities of real-world driving scenarios more effectively.