The final project of Umich ROB 498: Robot Learning for Planning and Control. In this project, we study the control of a double inverted pendulum on a cart(DIPC) system using different model based trajectory optimization algorithms - Model Predictive Path Integral (MPPI) and Differential Dynamic Programming (DDP).
To install the required environment and packages, we recommend to use Anaconda and run the following command:
conda env create -f environment.yml
conda activate venv
python main.pyThe following figure show the result of the inverted double pendulum system under random actions.
The following figure show the result of the inverted double pendulum system under MPPI. Our MPPI controller can stabilize the system when the pole angle is not too large.
The following figure show the result of the inverted double pendulum system under DDP. Stay tuned for a better result!
