Yeah, these are not good use cases for ML. They're tractable with traditional sensors and approaches, and they're horribly annoying when they get it wrong, but

Yeah, these are not good use cases for ML. They're tractable with traditional sensors and approaches, and they're horribly annoying when they get it wrong, but even more horribly annoying when they get it wrong inconsistently.