This is a solution to the CartPole-Swing Up problem using a pre-existing iLQR implementation from https://github.com/TGlad/ILQR.

• added a CartPole class / problem definition, following the example Acrobot class / problem definition

  • implemented the dynamics CartPole::f(), copying from the Python code CartPole.step() provided in the assignment
  • implemented the cost functions CartPole::l(), CartPole::lf(), finding a penalization scheme that produces the desired final state

• added cartpole_main.cpp that runs the pre-existing iLQR implementation / solver on the above CartPole problem

  • copied parameter values (time delta td, trajectory length T, initial state x0) from the assignment code

git diff wrt. fork base:

https://github.com/TGlad/ILQR/compare/22cf187bbc11d29de83eae57d413d0c3823aa687...wiktortomczak:ILQR:master

optimized

g++ cartpole_main.cpp -o cartpole_main -Wall -O3

debug

g++ cartpole_main.cpp -o cartpole_main -Wall -O0 -g

requires Eigen C++ library (Debian/Ubuntu: apt install libeigen3-dev)

$ cartpole_main                                # outputs cartpole.trajectory.csv
$ cartpole_render.py cartpole.trajectory.csv   # outputs cartpole.mp4

• cartpole.h - CartPole problem definition (dynamics & cost functions)
• cartpole_main.cpp - executable, runs iLQR implementation on the above problem
• cartpole.trajectory.csv - iLQR output, optimized (state, control) trajectory
• cartpole.mp4 - visualization of the above trajectory
• cartpole_render.py - visualization code, extracted from the assignment code

TODO: cartpole_main static binary