IGT-MPC: Implicit Game-Theoretic MPC

This repository contains the implementation of the paper "Learning Multi-agent Motion Planning Strategies from Generalized Nash Equilibrium for Model Predictive Control" submitted to 2025 Learning for Dynamics and Control Conference (L4DC)

Hansung Kim ([email protected])   Edward L. Zhu ([email protected])   Chang Seok Lim ([email protected])   Francesco Borrelli

Demonstration Video:

Clone repository

git clone https://github.com/MPC-Berkeley/Implicit-Game-Theoretic-MPC.git

Two vehicle Head-to-Head Racing

Install dependent libraries

cd head2head_racing/lib/mpclab_common
pip install -e .

cd head2head_racing/lib/mpclab_controllers
pip install -e .

cd head2head_racing/lib/mpclab_simulation
pip install -e .

The solver for MPC also requires hpipm_python package which you can download from here https://github.com/giaf/hpipm/tree/master

python evaluate.py

Two vehicle Un-signalized Intersection Navigation

Download the Game-Theoretic Interaction Dataset:

https://drive.google.com/drive/folders/1_8X7iMNEwCyPxwwrzvA_sD0aoYWLmUq4?usp=drive_link and unzip in 'intersection_navigation/game_theoretic_NN/dataset/' as shown below

├── intersection_navigation
  ├── game_theoretic_NN
    ├── configs
    │   ├── sc1_config.yaml
    │   ├── ...
    ├── dataset
    │   ├── instruction.txt
    │   ├── processed_sc1.pkl
    │   ├── processed_sc2.pkl
    │   ├── processed_sc3.pkl
    │   ├── processed_sc4.pkl
    │   ├── processed_sc5.pkl
    │   ├── processed_sc6.pkl
    │   ├── processed_sc7.pkl
    │   └── processed_sc8.pkl
    └── models
        ├── V_GT_sc1.pt
        ├── ...

python evaluate.py --save_dir <save_directory> --eval_mode <mode: str> --sc <int>

1. Replace <save_directory> with a local directory
2. eval_mode: [gt_mpc,mpc]
3. sc: [1,2,3,4,5,6,7,8]