You may have heard it said that flying an airplane is easy, the only hard part is landing it. Of course that is an oversimplification, but there is some truth in it, and the same holds true when it comes to flying drones. So as more and more drones take to the skies to deliver packages, inspect infrastructure, and help with rescue missions in the future, safer and more efficient methods to manage the landings of all of these vehicles will be in high demand.
Many of today’s best options take a two-pronged approach that utilizes both a traditional camera and a mmWave sensor located at the landing pad. The camera is used to determine the exact location of the drone over the landing pad, while the mmWave sensor provides information about exactly how far away it is. But these solutions all share a common problem — cameras have a much lower sampling rate than mmWave sensors, and that creates unacceptable bottlenecks in system throughput.
An overview of the system architecture (📷: H. Wang et al.)
A group led by researchers at Shenzhen International Graduate School and Tsinghua University have taken a somewhat different approach to drone landings that may help to overcome this issue. They have developed a new system called mmE-Loc that combines mmWave radar with event cameras, which are high-speed, bio-inspired sensors that respond to changes in pixel intensity with very little latency, as they capture only changes in scenes, not full frames. This change makes it possible for the cameras to keep up with the mmWave radar.
The mmE-Loc system uses two specialized processing modules to make the most of this data. First, the Consistency-Instructed Collaborative Tracking module filters out environmental noise and detects drones by exploiting the unique temporal consistency between the sensors and the micro-movements of the drone itself. This helps isolate the drone’s signals from background clutter. Then, the Graph-Informed Adaptive Joint Optimization module takes over, fusing data from both sensors using a factor-graph-based method to deliver highly accurate location estimates in real time.
The experimental setup (📷: H. Wang et al.)
In tests spanning over 30 hours of drone flights in both indoor and outdoor conditions, mmE-Loc outperformed four state-of-the-art localization systems. It achieved an average error of just 8.3 centimeters and latency of only 5.15 milliseconds. These results are over 48% better in accuracy and 62% faster in latency than existing methods.
Beyond lab testing, mmE-Loc was also deployed at a commercial drone delivery airport, where it demonstrated its robustness and practical feasibility in real-world operations. Even under variable lighting and environmental conditions, it maintained high precision and fast processing speeds.
The performance of the system, and its success in real-world deployments, signals that mmE-Loc might find a place in ensuring safe and efficient done landings in the years to come.