With Real-time Neural MPC you can efficiently integrate large, complex neural network architectures as dynamics models in an MPC-pipeline. Compared to prior implementations we can leverage neural networks with a 4000x larger parametric capacity in a 50Hz real-time framework.
ALT Quadrotor overflying a table in close proximity to the plane. Our approach can model the aerodynamic effects in close proximity to the ground, substantially limiting the horizontal tracking error.