Autonomous Navigation using mmWave Radar Sensors
This demo presents a real-time radar-only odometry and mapping system using four Texas Instruments IWR6843ISK mmWave radar demo boards. The system estimates the motion of a mobile robot using Doppler and angle information extracted from the radar point clouds, without relying on scan matching, cameras, or inertial sensors. Velocity is estimated directly from Doppler measurements and integrated over time to reconstruct the robot trajectory and map the surrounding environment. All processing is done on a central laptop running ROS2, with real-time visualization of both odometry and mapping in RViz. The radar setup is compact, self-contained, and designed for operation in GNSS-denied or visually degraded environments. This work highlights the unique potential of mmWave radar for ego-motion estimation and mapping, with robustness to clutter, lighting changes, and occlusions. The novelty lies in the use of Doppler-only odometry from multiple synchronized TI demo boards, without traditional scan matching or SLAM back-ends.
Project Objectives
- Demonstration of how Doppler-only odometry from multiple synchronized TI demo boards can be used to estimate the robot’s position
- Highlight the unique potential of mmWave radar for ego-motion estimation and mapping, with robustness to clutter, lighting changes, and occlusions.
- The system is designed for autonomous navigation applications by using Radar-based environmental mapping
Hardware & Software Units
- TurtleBot 4 Standard
- Four Texas Instruments IWR6843ISK mmWave radar
- An inertial measurement unit (IMU) Xsens MTi-7 Development kit
- Laptop running Ubuntu 22.04 and ROS2
Signal Processing
- Range-Doppler-angle cube computation
- for each sensor
- Doppler-based velocity estimation
- Odometry estimation via extended Kalman Filter (EKF) from velocity estimates
- Robot’s velocity is estimated by applying a Least Squares approach to the Doppler-shifted radar data and the corresponding point cloud measurements
Trajectory Mapping
OctoMap – An Efficient Probabilistic 3D Mapping Framework Based on Octrees
