In my opinion these fields will merge at some point in time as nextgen LLMs can be used to reason trajectories.

Consider RoboAgent[1] or LBMs at TRI[2]

[1] https://robopen.github.io/media/roboagent.pdf

[2] https://pressroom.toyota.com/toyota-research-institute-unvei...

I think from a technological perspective, building actual hardware has more relevant tangible real-world impact. But in the end, as many other posters have said, it really depends on the team, their experience and their capability to build product/drive business.

Shameless plug, I have been writing about this since 2023. [3] https://jdsemrau.substack.com/t/robotics