New robots open up new opportunities we never thought possible. The control design is intertwined with the nature of the system (rather than separate). My entry into this field was the Ionocraft.
Fun learning problem
They also are aligned very closely to my machine learning interests because of their high cost-per-test and unknown dynamics. When creating a new controller it becomes a minimum sample complexity problem — we want the robot to work as soon as possible, rather than being worried about peak performance after hours of training.
Too much uncertainty!
New types of robots suffer from uncertainty in manufacturing, assembly, and everywhere. The methods for using these robots must at least be able to handle these types of uncertainty, but ideally it will be able model and capture them individually.