Several months ago, I started building a framework that attempted to take advantage of the flexible nature of python, and recently got it to a point where it works pretty well in those areas. You can find it here: https://github.com/Team4819/4819-2014-Offseason-RobotPy . I showed it to virtuald, recently, and he pointed out several key issues with the design concept, as well as the fact that it works strikingly similar to the command framework, which I was unaware existed for python.
Faulty implementation aside, I still was able to do a few neat things with it:
Module reloading: In my framework, I grouped all of the code relevant to a subsystem of the robot into a single python file, which I call a module. These modules are dynamically imported into the framework in such a way that they can, later, be unloaded entirely, and then re-loaded, again. This allowed me to live-edit the individual modules and, with a button press, see results immediately on the robot, without even leaving teleoperated mode!
Error recovery: I used a config file to specify, for each module, a "fallback list" of alternative modules with the same basic abilities. Then, mid-run, if the loaded module threw an exception, it would be immediately unloaded and replaced by the next module on the fallback list, which could be a more basic and robust version.
I think that python as a FRC language as so much more potential than it is currently getting used for. If there were an official framework designed with python in mind, and it's abilities as an interpreted language, I think robotpy could become much more than just an alternative programming language for robots.