The Future of Vision and Sensitivity

We're drawing inspiration from some of the techniques that are used in LLMs and we're seeing where in our stack can we make it smarter because of these techniques. For example, instead of coming up with trajectories in a more classical motion planning way, can you use large models trained on lots of data to suggest trajectories without having to tell it how to generate a trajectory? And we're starting to do that. Also, simulation, I know I'm going on, but hopefully this is interesting to some people who think about these things. We can simulate raw sensor data. It's really nice to be able to have layers of simulation where you can say, OK, I'm going to