What Did You Find Out?
Radar odometry, the process of determining a robot's movement relative to a previous position using radar signals, has historically been limited by challenges in interpreting complex data. The key contribution of this work is a method to achieve exceptional precision in radar odometry with spinning radar sensors. Unlike traditional methods that rely solely on point-to-line comparisons between two instants of time, CFEAR integrates point-to-point comparisons from several consecutive radar scans to provide highly accurate odometry estimates, with the number of scans serving as a variable that can be adjusted to balance speed and accuracy.
This method allows robots to navigate even in environments with heavy smoke, dust, or fog — conditions that incapacitate LiDAR and cameras. Impressively, the algorithm achieved drift as low as 1.09%, setting a new benchmark for odometry performance.
What Have Been the Major Challenges?
Interpreting radar data is complex. Compared to cameras or LiDAR systems, radar generates vast amounts of data that are difficult to interpret since they can be affected by multiple reflections and noise. AI methods, typically used to process such data, often struggle to generalize across different environments. To address this, we opted for a more classical approach, which comes with the advantages of being very efficient, transferable and explainable.
What Are These Results Good For?
The research findings have several practical applications. In search and rescue missions, robots using the CFEAR system can navigate environments with poor visibility, such as burning buildings or disaster zones, helping to locate survivors or assess structural safety. In industrial and farming environments, radar-guided robots can operate efficiently, even in dusty conditions, such as those in factories or agricultural fields, where traditional sensors might struggle.
Future work aims to explore cost-effective radar technologies and the potential of Doppler radar, which measures distance and also detects the speed of moving objects. The open-source availability of the CFEAR algorithm invites researchers and industries worldwide to adapt and expand upon this promising approach.
Publication
Lidar-Level Localization With Radar? The CFEAR Approach to Accurate, Fast, and Robust Large-Scale Radar Odometry in Diverse Environments; Daniel Adolfsson, Martin Magnusson, Anas Alhashimi, Achim J. Lilienthal, Henrik Andreasson; IEEE Transactions on Robotics; 2024