A VLM (a vision-language model) is already being used by Waymo[1]. It's useful for scenarios that require reasoning and general knowledge.

[1] https://waymo.com/blog/2025/12/demonstrably-safe-ai-for-auto...