mmE-Loc: Facilitating Accurate Drone Landing with Ultra-High-Frequency Localization

TL;DR

mmE-Loc tackles the challenge of precise, low-latency drone landing localization in environments where GPS/RTK may fail by pairing an event camera with mmWave radar on the ground platform. It introduces two main components: CCT, which exploits temporal consistency and drone micro-motion/geometry for robust detection, and GAJO, which uses a factor-graph framework with motion-aware adaptive optimization to fuse modality data and produce accurate trajectories with ms-scale latency. The approach achieves cm-level accuracy and sub-10 ms latency in real-world tests, outperforming state-of-the-art baselines and demonstrating robustness to illumination, occlusion, and background motion. Its real-world deployment at a drone airport and low-resource efficiency suggest strong practical impact for scalable, safe low-altitude drone operations, offering a complementary localization pathway alongside RTK and visual markers.

Abstract

For precise, efficient, and safe drone landings, ground platforms should real-time, accurately locate descending drones and guide them to designated spots. While mmWave sensing combined with cameras improves localization accuracy, lower sampling frequency of traditional frame cameras compared to mmWave radar creates bottlenecks in system throughput. In this work, we upgrade traditional frame camera with event camera, a novel sensor that harmonizes in sampling frequency with mmWave radar within ground platform setup, and introduce mmE-Loc, a high-precision, low-latency ground localization system designed for precise drone landings. To fully exploit the \textit{temporal consistency} and \textit{spatial complementarity} between these two modalities, we propose two innovative modules: \textit{(i)} the Consistency-instructed Collaborative Tracking module, which further leverages the drone's physical knowledge of periodic micro-motions and structure for accurate measurements extraction, and \textit{(ii)} the Graph-informed Adaptive Joint Optimization module, which integrates drone motion information for efficient sensor fusion and drone localization. Real-world experiments conducted in landing scenarios with a drone delivery company demonstrate that mmE-Loc significantly outperforms state-of-the-art methods in both accuracy and latency.

Figures (30)

• Figure 1: Snapshot of drone landing phase, deliver drone airport, and performance of different sensors. (a) A delivery drone lands on the landing platform. (b) The real-world drone airport is equipped with multiple drones for package delivery. (c) Integrating mmWave radar with event camera combines reliable depth sensing and 2D imaging at ultra-high sampling frequencies. This solution achieves high spatio-temporal resolution 2D sensing and precise depth sensing for drone ground localization, while maintaining full compatibility with flight controllers by operating at update frequencies exceeding 150Hz.
• Figure 2: Benchmark study on drone localization. (a) We conduct the benchmark study at a real-world drone delivery airport. (b) Both the event camera and mmWave radar are sensitive to environmental variations, leading to significant sensing noise. Within a $1~\text{ms}$ interval, event camera background produces $764$ noise events ($25.3\%$ of total), while mmWave radar background yields $22$ noise points ($71.5\%$ of total). (c) Comparison of noise filtering performance in terms of Recall and Precision. The filtering rate is defined as $\text{Recall} = \frac{N_{\text{filtered, true}}}{N_{\text{noisy}}}$ and $\text{Precision} = \frac{N_{\text{filtered, true}}}{N_{\text{filtered, total}}}$, where $N_{\text{filtered, true}}$ is the number of correctly removed noise samples. A value of $100\%$ indicates the ideal case where all noise samples are correctly removed (Recall = 100%) and no valid samples are mistakenly removed (Precision = 100%). Existing algorithms rely solely on either mmWave radar or event cameras, failing to exploit their cross-modal correlations, which limits their noise filtering capability. (d) These algorithms are prone to cumulative drift errors and experience considerable delays.
• Figure 3:
• Figure 4: Distance calculation by frequency difference
• Figure 5: Direction calculation by phase difference