DCMRTA

Code for ICRA2024 Paper: Dynamic Coalition Formation and Routing for Multirobot Task Allocation via Reinforcement Learning.

This is a repository using deep reinforcement learning to address single-task agent (ST) multi-robot task(MR) task assignment problem. We train agents make decisions sequentially, and then they are able to choose task in a decentralized manner in execution.

Demo

Code structure

Three main structures of the code are as below:

1. Environments: generate random tasks locations/ requirements and agents with their depot.
2. Neural network: network based on attention in Pytorch
3. Ray framework: REINFORCE algorithm implementation in ray.

Running instructions

1. Set hyperparameters in parameters.py then run python driver.py

2. How to test:

   1. Generate a new test set by choosing environment parameters in TestSetGenerator.py then run python TestSetGenerator.py
   2. OR-Tools: Change parameters in baselines/OR-Tools.py then run python baselines/OR-Tools.py.
   3. CTAS-D: see original paper for details

      1. cd baselines/CTAS-D
         mkdir build && cd build
         cmake ..
         make

      2. then run bash CTAS-D_test.bash
   4. get results from python results_plotting.py and the results will be save in tesetSet/metrics

3. requirements:

   1. python => 3.6
   2. torch >= 1.8.1
   3. numpy, ray, matplotlib, scipy, pandas, ortools