Winding-Number-Aware Navigation

This repository contains the code for the paper "Symmetry-Breaking in Multi-Agent Navigation: Winding Number-Aware MPC with a Learned Topological Strategy".

The framework of our codes is based on the work (sriyash421/Pred2Nav), and the codes for CADRL are based on the work (vita-epfl/CrowdNav).

This sim_iros branch contains the code for the holonomic simulation experiment. For the code used in the real-robot experiment, please see maru_exp_iros branch of this repository.

Tested Environment

• Manjaro Linux 24.0.4
• Python 3.11.9

Installation

# create venv
python -m venv .venv
source .venv/bin/activate

# install requirements
pip install -r requirements.txt
pip install git+https://github.com/sybrenstuvel/Python-RVO2.git  # for ORCA

Getting started

if you want to run other policy, you can change the robot_policy in the config/experiment_param.yaml file.

• available policies: ORCA, CADRL, VanillaMPC, MeanMPC (T-MPC), WNumMPC (proposed method),
  • default: WNumMPC
• Agent Counts: 3, 5, 7, 9
  • default: 9
• Placement Generation: random (gen), crossing (opp)

As a specific rule of placement: if Agent count = 9 and placement generation = gen, set eval_states: ./datas/eval_states/8-100-gen.npy.

Run WNumMPC Policy

To execute the trained policy, run the following script. If you want to change the settings, modify experiment_param.yaml.

python eval_policy.py

Training WNumMPC Policy

To run the policy training, execute the following script. If you want to change the settings, modify experiment_param.yaml and h32/h64.yaml.

python train_ppo.py

Citation

@misc{nakao2026symmetrybreakingmultiagentnavigationwinding,
      title={Symmetry-Breaking in Multi-Agent Navigation: Winding Number-Aware MPC with a Learned Topological Strategy}, 
      author={Tomoki Nakao and Kazumi Kasaura and Tadashi Kozuno},
      year={2026},
      eprint={2511.15239},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2511.15239}, 
}