First, and most important obviously, this is pretty neat research into training robot motion with online, and non-optimal, feedback. It’s a nice consideration of the type of feedback one is likely to get, or to get easily. And the illustrative video on their page is pretty great (I found the moment when they showed the robot how to point the knife towards itself, not someone else, adorable…)
But it’s also worth noting that the TechCrunch story on the research is pretty hysterical: “Cornell Researchers Help Robot Unlearn Stabby Motions With A Human Trainer”.