This repo contains code related to vehicle dynamics written by Justin Yurkanin ([email protected]) The different branches refer to different experiments. float_model uses Robcogen templated with float scalar type and is fastest and should be used for path planning and anything serious. bekker_descent uses an auto-diff scalar type for doing gradient descent to identify bekker parameters. hybrid_network is similar and trains a physics based neural ODE. 2d_neural_ode is a test branch for training 2D neural ode's for vehicle dynamics. adjoint_method is another test branch to see if the adjoint sensitivity method works better for neural ODE's than just directly calculating the parameter/loss gradient with auto diff.
This repo requires CppAD for auto-diff, libeigen3, and Robcogen for forward dynamics.
For a more in depth description of what all this code does, see my thesis.