DR-DDP

This repository containts the official Python implementation of DR-DDP algorithm presented in the paper Distributionally Robust Differential Dynamic Programming with Wasserstein Distance.

Requirements

To run the code in this repository, make sure you have the following dependencies installed:

• Python (>= 3.6)
• Gym
• PyTorch
• SciPy
• Other standard Python libraries, such as NumPy, Matplotlib, and Pickle.

Installation

Follow these steps to install DR-DDP and its dependencies:

1. Clone the repository:

$ git clone https://github.com/CORE-SNU/DR-DDP.git

2. Navigate to the DDP Systems folder:

$ cd DR-DDP/DDP_Systems

3. Install the Gym environments:

$ pip install -e .

Usage

In the main DR-DDP directory, run the main.py code with the desired arguments:

$ python main.py --dist <distribution> --num_sim <num_simulations> --num_samples <num_samples> --system <system_type> --baseline <baseline_type> --plot --render

Here's an explanation of each argument:

• --dist <distribution>: Specify the disturbance distribution to use. Replace <distribution> with the desired distribution name (either normal or uniform).
• --num_sim <num_simulations>: Set the number of simulations to run. Replace <num_simulations> with the desired number.
• --num_samples <num_samples>: Set the number of samples to generate in each simulation. Replace <num_samples> with the desired number.
• --system <system_type>: Specify the system to be controller. Replace <system_type> with the desired system name (either Oscillator-v0 or Car-v0)
• --baseline <baseline_type>: Specify the type of baseline controller. Replace <baseline_type> with the desired baseline type (1 to use GT-DDP, 2 to use both GT-DDP and box-DDP).
• --plot: Include this flag to enable plotting of the simulation results.
• --render: Include this flag to enable rendering of the system.

Example

To apply the DR-DDP algorithm for a kinematic car navigation problem, run the following command:

$ python main.py --dist uniform --num_sim 1 --num_samples 10 --system Car-v0 --baselin 2 --plot

The following figure demonstrates an example of the resulting trajectories: