Polarquest - ROS-Based System Identification & Model Validation for Autonomous Vehicle

A compact ROS stack to use data drive system identification for the Polaris GEM vehicle, publish paths from CSV files, and run a Model Predictive Controller (MPC) built from the identified vehicle dynamics.
Everything runs inside a Docker environment with optional NVIDIA GPU acceleration.

---

Overview

This project provides a self-contained setup to:

• Identify the Polaris GEM vehicle dynamics from recorded data.
• Build and validate an MPC controller based on the identified model.
• Run and visualize the full system in Gazebo + RViz.
• Publish custom reference paths for tracking experiments.

All mathematical analysis, system identification steps, and controller derivations are detailed in
analysis/system_identification.ipynb.
A rendered PDF version is also provided for convenience:
analysis/system_identification.pdf.

---

Why This Project Matters?

This project implements a full ROS-based system identification and validation pipeline for a Polaris GEM vehicle. The goal is to:

• Collect vehicle data using ROS
• Identify a physics-consistent dynamic model
• Validate the model against real-world recordings
• Quantify performance using cross-track error (CTE)
• Prepare the model for MPC-based trajectory tracking

System Architecture

flowchart LR
    subgraph ROS Ecosystem
        A[Ackermann Msg]
        B[Odom]
        D[rosbag Recording]
    end

    subgraph ROS Ecosystem
        F[MPC Controller]
        G[Ackermann Msg]
    end

    A --> D
    B --> D

    D --> E[System Identification Notebook]
    E --|Identified Paramenters|--> F
    
    F --> G
    G --> H[Cross-Track Error Analysis]

Loading

---

Validation Results

Model vs Real Pose Tracking (x, y, ψ)

The identified model closely follows the recorded vehicle dynamics.

Cross-Track Error (CTE)

Final RMSE: 0.337 m

Installation (Docker + NVIDIA container toolkit)

Follow the official guides:

Docker: https://docs.docker.com/engine/install/ubuntu/

Nvidia Container Toolkit: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html

After installation:

# Bring up the preconfigured ROS Noetic container
cd <project-root>/docker
docker compose up -d noetic

# Open a shell inside the workspace
docker exec -it -w /home/dev/code/polarquest/ws noetic-dev bash

Quick Start

Launch everything

You can start the full simulation and MPC controller with:

roslaunch polaris_system_id gem_mpc_follower.launch

Or use the helper script, which launches everything and displays the tracking error plots when stopped:

./src/polaris_system_id/launch/launch_and_show_rmse.sh

Manual workflow

# 1) Gazebo + RViz
roslaunch gem_gazebo gem_gazebo_rviz.launch

# 2) Odometry → TF bridge
rosrun polaris_system_id odom_to_tf.py

# 3) Path from CSV
rosrun polaris_system_id publish_from_csv.py

# 4) MPC controller
#    steer_only: steering control with constant speed (_v_fixed)
rosrun polaris_system_id mpc.py _mode:=steer_only _v_fixed:=6.0 _Ts:=0.05 _Np:=20 _solve_every:=3
# or
#    speed_steer: steering + speed control
rosrun polaris_system_id mpc.py _mode:=speed_steer _Ts:=0.05 _Np:=20 _solve_every:=3 _spd_min:=0.5 _spd_max:=8.0 _spd_slew:=0.6