I recently finished teaching an optimization-based controls course and found a but of small things in the code that needing fixing/improving:

• Parameters were not being passed properly to differentially flat systems.
• Basis functions derived from BasisFamily were not calling BasisFamily.__init__.
• Changed constraints keyword in optimal.solve_ocp and flatsys.point_to_point to trajectory_constraints for more consistency (legacy keyword still works).
• Initial conditions for simulation of discrete time systems was being passed as a list instead of an ndarray, which caused some odd behavior and wasn't consistent with continuous time system usage.
• Added check for constraint types in optimal control problem setup (rather than when solving the optimal control problem).
• Fixed small inconsistency in the way discrete time was handled in setting up an I/O system for model predictive control (should be inherited from system, not passed separately).
• Unused keywords in OptimalControlProblem now generate an error (mentioned here).
• To be consistent with the normal problem formulation, the discrete time optimal cost calculation no longer includes the endpoint (use terminal_cost instead).
• create_mpc_iosystem now re-computes the optimal controller when returning the output, which eliminates an unintended one-step delay in the MPC dynamics.
• Added unit tests.

Coverage increased (+0.006%) to 93.601% when pulling f807b33 on murrayrm:cds112 into e10855e on python-control:master.