I spent a while getting really frustrated with this exercise before realizing what was wrong. The tests use defvar to create the robots that the rest of the tests use, meaning that the robots only get created once the first time you run the tests. I ended up having to run them from the terminal instead so I could isolate each run, but I think switching that to setq instead would solve the issue without having to refactor the tests too much.
This particular exercise is in the Summer of Sexps, so it would be really nice not to have this road-bump.