Physics-Informed Neural Nets for Control (PINCs) extend traditional Physics-Informed Neural Networks (PINNs) by incorporating control inputs and variable initial conditions, therefore enabling autoregressive long-horizon simulation by chaining short-horizon predictions, and making them suitable for control-oriented applications. However, the existing literature commonly adopts PINC training procedures where the governing dynamics are embedded in an explicit analytical form, whereas in many engineering workflows, the system models are often available only as Functional Mock-up Units (FMUs). This paper proposes an FMU-based training framework for PINC that uses an FMI 3.0 Model Exchange (ME) FMU directly as the description of system dynamics. The physics-informed residual is evaluated in the forward pass by querying the FMU for state derivatives, while gradient backpropagation is enabled through FMI 3.0 adjoint derivatives integrated into PyTorch via a custom autograd.Function. Case studies on a Van der Pol oscillator and on a mechanical system show that the proposed framework can train PINCs without manually implementing the analytical governing equations and can achieve long-horizon simulation performance comparable to conventional ODE-based training when the required FMI 3.0 features are available.