Skip to content

Latest commit

 

History

History

simple_pendulum_system

ReadMe.md

Instructions

In this study, we investigate the use of Pontryagin Differential Programming (PDP) towards performing Inverse Optimal Control (IOC) of discrete-time, nonlinear systems. In particular, for given system dynamics with control objective parameterized by weights w, we consider the problem of estab- lishing optimal values of the weights w, minimizing the imitation loss L between the resulting system trajectories and some desired system trajectory. In order to find such weight, we use the PDP methodology to compute the gradient dL dw of the loss function with respect to the weights. Using this gradient, we can then apply a steepest descent algorithm to compute improved estimates of the weights, iteratively converging to a local minimizer of the loss function. The proposed methodology is examined across a diverse range of dynamic systems, verifying the results for a cart- pole system presented in earlier works, as well as successfully applying the method to perform IOC of an inverted pendulum and a quadrotor system.

Requirements

This project depends on the python modules numpy, scipy, matplotlib, casadi, ffmpeg-python.

pip install numpy scipy matplotlib casadi ffmpeg-python

Initial Step

Go to file 'run_pdp_user_interface.py' and change the parameter variable 'user_choice' from 1 to 5 successively to go through successive stages of computing pdp loss and achieving inverse optimal control.

Steps to execute the sequence of PDP (you can skip step 2 and move on to 3, 4, 5 for viewing obtained results if needed)

  1. user_choice = 1 : Generate optimal trajectories with respect to assumed quadratic cost (Primal Optimization Program).
  2. user_choice = 2 : pdp iteration (Computing PDP loss, weight updates using Auxillary Optimization Program and Differential PMP).
  3. user_choice = 3 : plot loss trace vs iteration.
  4. user_choice = 4 : plot trajectories vs ground truth, plot weights vs iteration.
  5. user_choice = 5 : generate animations for compare primal optimizaton program and auxillary optimization program.

Warnings:

While switching from one system model to another model, please restart the kernel or clear the variables of the previously used system model and run it step wise with above choices in succession to avoid runtime errors.