Robot Localization

This project was developed as part of MEEN 700: Robotic Perception

This project implements on-board localization for a mobile robot using data collected from a smartphone mounted on the robot platform. The localization system leverages an Extended Kalman Filter (EKF) to estimate the robot's 2D position and orientation in real-time using fused IMU (accelerometer and gyroscope) and GPS data. This approach provides a lightweight, quick-to-deploy alternative to traditional ROS-based localization pipelines.

---

Features

• Sensor fusion using EKF with an 8-state vector: [px, py, vx, vy, ψ, bax, bay, bwz].
• Input from smartphone sensors via the Sensor Logger app.
• Conversion of IMU data from phone frame to robot body frame using rotation matrices.
• Automatic trimming of stationary periods and bias estimation from static IMU readings.
• Visualization and benchmarking using ground-truth comparisons (if available).

---

System Architecture

The system includes:

• Python-based EKF pipeline for prediction and correction.
• Preprocessing for sensor calibration and frame alignment.
• Visualization for plotting estimated trajectories.

Folder Structure

├── main.py              # Entry point for executing the pipeline
├── data_loader.py       # CSV loading and timestamp alignment
├── preprocessor.py      # Static bias estimation, trimming, calibration
├── utils.py             # Frame transformation and GPS to ENU conversion
├── ekf_filter.py        # Core EKF logic (prediction + GPS update)
├── tests/               # Sample datasets with IMU + GPS
├── phone_localization/  # Output CSVs (results)

---

Usage

1. Prepare Data

Place your sensor logs (CSV format) in the tests/ directory. Files must include:

• Accelerometer and gyroscope (100 Hz)
• Orientation (yaw)
• GPS (lat, lon, horizontal accuracy, 1 Hz)

2. Run Localization

Run the following command to execute the pipeline:

python main.py

3. Review Results

Output files will be generated in the phone_localization/ directory:

• ekf_states.csv:

  • px, py: Estimated positions
  • vx, vy: Estimated velocities
  • ψ: Orientation (yaw)
  • bax, bay, bwz: Estimated biases

• Plots: Visualized trajectories for validation.

---

Results & Benchmarks

• Estimated paths closely match actual movement across tested configurations.
• Robust to GPS dropout (1 Hz) using high-frequency IMU integration.
• Sensor alignment and pre-processing significantly improve accuracy.

---

Known Issues

• GPS Cold Start: TTFF (time to first fix) may exceed one hour on new modules.
• Magnetic Interference: Phone compass can be unreliable near motors/LiDAR.
• Mount Stability: Wind and vibration can affect sensor readings.
• Frame Alignment: Misaligned frames reduce accuracy if not corrected properly.

---

Future Improvements

• Adaptive tuning of EKF noise models for varying environments.
• Integration with ROS for real-time robotic deployments.
• Extended support for visual odometry or wheel encoders.

---

Contributors

• Abdulaziz Alharbi – Hardware setup, data acquisition, and documentation
• Ammar Waheed – System design, EKF implementation, and documentation
• Hassan Niaz – Mathematical modeling and EKF development

---

License

This project is licensed under the MIT License. See the LICENSE file for more details.