CSC-MPPI: A Novel Constrained MPPI Framework with DBSCAN for Reliable Obstacle Avoidance

This repository provides a JAX-JIT accelerated implementation of the Constrained Sampling Cluster MPPI (CSC-MPPI) algorithm.
CSC-MPPI introduces a novel constrained sampling framework based on DBSCAN clustering and primal-dual gradient optimization to improve obstacle avoidance and constraint satisfaction in sampling-based Model Predictive Path Integral (MPPI) control.

The algorithm is designed for high-performance execution on GPU using JAX with just-in-time (JIT) compilation, enabling fast and parallelized trajectory rollouts.

---

📰 Publication

📄 Title: CSC-MPPI: A Novel Constrained MPPI Framework with DBSCAN for Reliable Obstacle Avoidance
🛠 Authors: Leesai Park¹, Keunwoo Jang²†, and Sanghyun Kim¹³†
📅 Conference: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2025

• 🔗 Project Website: https://cscmppi.github.io/
• 📹 YouTube Video: Watch Video

---

⚙️ Prerequisites

• Python 3.10+
• JAX with GPU support
• CUDA 12.1+
• Recommended: Use a virtual environment (e.g., conda)

✅ Create Virtual Environment

Please refer to the official JAX installation guide for detailed setup instructions compatible with your system and CUDA version.

conda create -n csc_mppi python=3.10
conda activate csc_mppi

conda install matplotlib
pip install -U "jax[cuda12]"
conda install -c rapidsai -c nvidia -c conda-forge -c defaults  cuml

---

🚀 Installation and Run

git clone <repository_url>
cd <repository_name>
python3 main/env1.py or main/env2.py

---

🎬 Experiment Results

Environment 1

CSC-MPPI	Standard MPPI

---

Environment 2

CSC-MPPI	CSC-MPPI wo DBSCAN

---