This is the source code of our paper A Fuel-Optimal Landing Guidance and Inverse Kinematics Coupled PID Control Solution for Power-Descent Vertical Landing in Simulation, a copy of the paper with typo corrections is also attached in this repo under <root>/the_paper/.
Regardless of many TODO items still pending in the current implementation, we decided to go open source aiming at helping people (especially students) get an easy start on this topic -- you can refer to this repo for setting up the simulation environment and our algorithm for basic closed-loop control implementation :)
If you're a student or lab person who's working on a similar topic at a very early stage, welcome to drop me a message at genxium@outlook.com for discussion.
Happy simulating!
Please ensure that you have python version >=3.8 but <3.11 installed (either on OSX or Windows, this is to ensure that we have same behaviour for scipy-1.7.1 solvers).
pip3 install -r requirements.txt
python3 lunarlander_pid_only.py
To export math typesetting in draw.io files correctly, please ensure that you have draw.io-16.0.2 or later installed.
Kindly note that you might encounter cases where the pid only controller doesn't work to land the lander. The feasibility of pid only approach is highly impacted by the Tsettle, because it's bound directly for the target position. If Tsettle were not estimated near a possible value the controller would've output impossible values to achieve and prohibited by plant clipping. Therefore a change on env.SCALE or start position could make pid only infeasible.
The gfold approach on the other hand only needs an "upper bound for Tsettle" and is very flexible for change in env.SCALE or start position.
It's also noticeable that a change of the env.SCALE has much stronger impact on Iz ~ m*(R^2) than m due to constant density of the lander, thus might require some tuning of rho_1, rho_2, rho_side_1 and rho_side_2 after such change.