c-ai-s / ml-casadi

Use PyTorch Models with CasADi and Acados

ML-CasADi

This is the underlying framework enabling Neural-MPC in our paper

Neural-MPC: Deep Learning Model Predictive Control for Quadrotors and Agile Robotic Platforms

Arxiv Link

This framework enables PyTorch Models to be used as CasADi functions and subsequently in Acados optimal control problems.

Examples

Arbitrary PyTorch Model as first- or second order approximation

import ml_casadi.torch as mc
import casadi as cs

model = mc.TorchMLCasadiModuleWrapper(
    torch_module,
    input_size=size_in,
    output_size=size_out)
    
casadi_sym_inp = cs.MX.sym('inp', size_in)
casadi_sym_out = model.approx(casadi_sym_inp, order=1)  # order=2

casadi_func = cs.Function('model_approx_wrapper',
                          [casadi_sym_inp, model.sym_approx_params(order=1)],
                          [casadi_sym_out])

casadi_param = model.approx_params(input, order=1)  # order=2
casadi_out = casadi_func(input, casadi_param)

Using Acados, the approximation parameters can be passed to the optimal control problem via acados_ocp_solver.set(n, 'p', casadi_param).

Specific PyTorch Model without approximation

import ml_casadi.torch as mc
import casadi as cs

model = mc.MultiLayerPerceptron(
    input_size=size_in,
    hidden_size=hidden_size,
    output_size=size_out,
    n_hidden=n_hidden,
    activation='relu')
    
casadi_sym_inp = cs.MX.sym('inp', size_in)
casadi_sym_out = model(casadi_sym_inp)

casadi_func = cs.Function('model_approx_wrapper',
                          [casadi_sym_inp],
                          [casadi_sym_out])

casadi_out = casadi_func(input, casadi_param)

Citing

If you use our work please cite our paper

@article{salzmann2022neural,
  title={Neural-MPC: Deep Learning Model Predictive Control for Quadrotors and Agile Robotic Platforms},
  author={Salzmann, Tim and Kaufmann, Elia and Pavone, Marco and Scaramuzza, Davide and Ryll, Markus},
  journal={arXiv preprint arXiv:2203.07747},
  year={2022}
}