SLAM: Navigation Stack From Scratch

Designed for the Turtlebot3, this project implements a full-stack SLAM architecture using an Extended Kalman Filter (EKF). The system integrates sensing, state estimation, and mapping to enable real-time localization in an unknown environment. The focus of this project was to develop a deeper understanding of robotics perception and navigation at the system control level.

Github Link: https://github.com/ncknight-un/SLAM_EKF_SCRATCH_TURTLEBOT3

SLAM-EKF in Simulation

• Comparison Setup: SLAM-EKF and odometry are simulated concurrently to highlight error accumulation in state estimation.
  • Scenarios: Teleoperation and circular trajectory tests
  • Method: SLAM-EKF vs. odometry
  • Result: EKF-based SLAM significantly reduces drift compared to odometry alone
  • Takeaway: Sensor fusion improves localization accuracy and robustness over time

Physical Testing on Turtlebot3

• When evaluated on a simple circular trajectory, the localization error was approximately 0.25 m in both the x and y directions.
  • This suggests that real-world performance is expected to exhibit greater error accumulation due to factors such as wheel slip, friction, and unmodeled dynamics.
• Ongoing work focuses on addressing the simulation-to-real gap for SLAM-EKF deployment on the physical robot.
  • Currently adjusting parameters for algorithms to better match noise consistancy with existing LiDAR senser and control nodes.

Sensing & Perception

The simulation in this project compares two state estimation approaches—raw odometry and SLAM with an Extended Kalman Filter (EKF)—to demonstrate the improvements in localization accuracy achieved through sensor fusion.

• LiDAR: Provides distance measurements and enables environmental perception for SLAM-based state estimation
• Odometry: Tracks wheel-based motion estimates, serving as a baseline for comparison

Sensor data is inherently noisy and uncertain, making filtering and probabilistic inference essential for reliable state estimation.

State Estimation (EKF)

For the SLAM state estimation, the Extended Kalman Filter is used to estimate the robot’s state, including position and orientation.

Key Steps:

• Prediction: Uses motion model to estimate next state
• Update: Corrects estimate using sensor measurements

The EKF enables robust localization by combining noisy sensor inputs into a consistent state estimate.

Acknowledgements

Thank you to Dr. Matthew Elwin, director of MSR at Northwestern, who teaches this project course (ME495: Sensing, Navigation, and Machine Learning).

Skills Improved

• ROS 2 development in C++
• CMake-based project architecture
• Full-stack robotics system design
• Robot navigation principles
• Sensor integration and control algorithms

Key Takeaway

This project demonstrates the importance of sensor fusion in achieving accurate and reliable localization, highlighting how SLAM-EKF significantly reduces drift compared to raw odometry in robotic systems.