EM-Patroller

Introduction

• This is the code repository for the paper "Entropy Maximized Multi-Robot Patrolling with Steady State Distribution Approximation" that was submitted to ICRA 2023.
• This is a joint research program among A*STAR, MIT and NUS. (© 2022 A*STAR. All rights reserved.)

Dependencies

• Python 3.8
• NetworkX 2.8.4
• Numpy
• Matplotlib
• PyTorch (GPU Acceleration is recommended)
• tqdm

Description

• We integrate our simulation toolbox into a Python package called emp.
• opt.py realizes the centralized optimizer for given environment and MRS.
• env.py constructs environments. Predefined topological graphs include HOUSE, OFFICE and MUSEUM.
• robot.py defines the structure and actions of individuals. Robots are fully decoupled from policies.
• policy.py implements parameterized policies, which are MLP in our case.
• main.py claims legal arguments of command line input and the procedure of our algorithm.

Usage

If you want to optimize a MRS of 4 robots in the MUSEUM environment with lr=1e-3 for 5,000 epochs and test it for 1,000,000 steps and require real-time feedbacks and final outputs, the command line would be like:

python main.py --env MUSEUM --size_MRS 4 --epoch 5000 --step 1000000 --lr 1e-3 --verbose --output

For more operations, please refer to parse_opt() in main.py.