@PPuchaud@TimSalzmann Optimal control/model predictive control with a learned system model, for example. There have been several couplings like this with TensorFlow and PyTorch before, but this is the first implementation that supports code-generation and therefore is potentially overhead-free.
CasADi demo https://t.co/zcHwfxKmtG - feel like deep-diving in a hands-on workshop? Early bird registrations for https://t.co/23YmChLy4P are still open for two weeks.
We propose a computationally-efficient Convex Lie Algebraic MPC for marine vehicles. The higher water density leads to significant environmental forces and state-dependent vehicle models. The proposed MPC significantly reduces computation time, enabling real-time implementation with longer planning horizons.
Paper: https://t.co/vYVYuGwa1m
Code: https://t.co/Rin92iaftH
CasADi 3.6.0 is here! Now compatible with new versions of #Python/#Octave/#Matlab & M1 hardware. New solvers: SPRAL, HPIPM, ProxQP & HiGHS. #FMI/#Modelica interoperability, improvements to ODE/DAE integrators, NLP solvers & symbolics. Details at https://t.co/OfFmyKbQGy
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.
First set of simulations: takes ~10 months to converge w/ good results
Second set of simulations on new data: takes ~12 hours to converge w/ great results on a first pass. Call me the CasADI master now.
I'm taking that as a win!! @JohnRHutchinson #DAWNDINOS