Teaching CartPole

The repository includes single and double cartpole environments, starter code for linear quadratic regulator and interface to control cartpole envs.

Dynamics data

Dynamics data can be accessed through this onedrive link. Updated versions of the data will be available with prior notice.

Note that you can also collect your customized data with the instructions provided below.

CLI and main entry

The main simulation file is cartpole_sim.py. You can use python to run the file with the described CLI args.

python3 cartpole_sim.py --env CartPole

The data collection mode can be activated by the following command for a single cartpole env.

python3 cartpole_sim.py --env CartPole --data-collection --dataset-name cartpole-dynamics-data

More CLI options are available if needed.

  --env ENV             Switch between single and double cartpole envs: CartPole, DoubleCartPole.

  --refresh-rate REFRESH_RATE
                        GUI refresh rate.

  --control-rate CONTROL_RATE
                        Control input will be applied once per timestep by default!

  --data-collection     Log data in a CSV file, later to be used for learning dynamics and more! This will override the control inputs.

  --sampling-rate SAMPLING_RATE
                        Samples are taken at each step by default.

  --reset-rate RESET_RATE
                        Resets the environment if data collection mode is active.

  --dataset-size DATASET_SIZE
                        Size of the collected dataset.

  --control-min CONTROL_MIN
                        Minimum control input for sampling from a uniform distribution.

  --control-max CONTROL_MAX
                        Maximum control input for sampling from a uniform distribution.

  --dataset-name DATASET_NAME
                        Pick a name for the collected dataset.