A robust baro-radar-inertial odometry m-estimator for multicopter navigation in cities and forests (2408.05764v1)

Abstract: Search and rescue operations require mobile robots to navigate unstructured indoor and outdoor environments. In particular, actively stabilized multirotor drones need precise movement data to balance and avoid obstacles. Combining radial velocities from on-chip radar with MEMS inertial sensing has proven to provide robust, lightweight, and consistent state estimation, even in visually or geometrically degraded environments. Statistical tests robustify these estimators against radar outliers. However, available work with binary outlier filters lacks adaptability to various hardware setups and environments. Other work has predominantly been tested in handheld static environments or automotive contexts. This work introduces a robust baro-radar-inertial odometry (BRIO) m-estimator for quadcopter flights in typical GNSS-denied scenarios. Extensive real-world closed-loop flights in cities and forests demonstrate robustness to moving objects and ghost targets, maintaining a consistent performance with 0.5 % to 3.2 % drift per distance traveled. Benchmarks on public datasets validate the system's generalizability. The code, dataset, and video are available at https://github.com/ethz-asl/rio.

• The paper demonstrates that the BRIO m-estimator effectively fuses FMCW radar, IMU, and barometric data for precise state estimation in GNSS-denied settings.
• It employs statistical tests and m-estimation techniques to suppress noise and outliers, overcoming challenges from visual degradation and unpredictable environments.
• Real-world tests in cities and forests validate its performance, maintaining drift rates between 0.5% and 3.2% per distance traveled.

A Robust Baro-Radar-Inertial Odometry M-Estimator for Multicopter Navigation in Cities and Forests

The research detailed in the paper presents an innovative approach to enhancing the navigation capabilities of multicopter drones operating in challenging environments, particularly those lacking reliable GNSS signals. The paper introduces a robust Baro-Radar-Inertial Odometry (BRIO) m-estimator designed to provide accurate and reliable state estimation for multirotor drones navigating through urban and natural settings, such as cities and forests. This method addresses the limitations posed by visual degradation, geometric ambiguity, and adverse environmental conditions that impair traditional navigation systems relying on vision, lidar, or thermal sensing.

Technical Approach and Methodology

The BRIO m-estimator integrates data from frequency-modulated continuous-wave (FMCW) radar, inertial measurement units (IMU), and barometric sensors to estimate the drone's position, velocity, and orientation. This multimodal sensing strategy capitalizes on the complementary strengths of radar and IMU technologies, where radar provides robust linear displacement measurements even in visually degraded conditions, and IMUs offer reliable orientation changes. The inclusion of barometric data further mitigates altitude drift, enhancing vertical state estimation.

By employing statistical tests within the radar-inertial odometry (RIO) framework, the estimator effectively suppresses noise and outliers arising from radar detections of dynamic and ghost objects. The use of m-estimation techniques ensures heightened robustness and requires minimal tuning compared to binary outlier rejection methods like RANSAC.

Empirical Evaluation

The paper reports extensive real-world tests conducted in both urban and forest environments. The findings demonstrate the capability of the BRIO system to maintain low drift rates of 0.5% to 3.2% per distance traveled, even in the presence of moving objects or adverse conditions that typically challenge other navigation systems. The validation benchmarks against public datasets further corroborate the system's generalizability to various configurations and scenarios.

Implications and Future Directions

This research contributes significant improvements to UAV navigation in GNSS-denied environments. The robustness of BRIO in dynamic and degraded conditions positions it as a viable alternative to lidar or vision-based systems. The modular nature and the reduced number of parameters that require tuning enhance the practicality of deployment across different multicopter platforms.

Future developments could explore the integration of additional environmental sensing modalities to further enhance system accuracy and robustness. Exploration into machine learning techniques for dynamic parameter adaptation might also offer avenues to refine the estimator's adaptability across diverse environmental conditions.

This work advances the field of UAV navigation by providing an effective and reliable state estimation solution, particularly beneficial for applications such as search and rescue operations, where navigating unstructured environments accurately and reliably is crucial.