Obstacle Avoidance onboard MAVs using a FMCW RADAR

Nikhil Wessendorp, Raoul Dinaux, Julien Dupeyroux, Guido de CroonIROS 2021 · OpenReview

TL;DR

This paper explores the use of miniature FMCW radars for obstacle avoidance on Micro Air Vehicles (MAVs) in challenging environments where traditional sensors like cameras fail. By implementing a robust multi-target tracking and avoidance pipeline, they demonstrate that a standalone radar can enable safe sense-and-avoid capabilities for MAVs, even in cluttered and low-visibility conditions.

What Problem Does This Paper Address?

Micro Air Vehicles (MAVs) are increasingly deployed for critical tasks in complex and hazardous environments, such as search and rescue in smoke-filled buildings or surveillance in dusty conditions. A key challenge for MAVs operating in these scenarios is reliable navigation and obstacle avoidance, especially when visibility is poor. While cameras are common, they falter in dust, fog, or smoke. Other sensors like ultrasound have limitations with soft or curved edges. Radar sensors, traditionally bulky and expensive, have recently become available in miniature FMCW (Frequency-Modulated Continuous-Wave) formats. However, their application for onboard MAV obstacle avoidance, particularly in indoor, cluttered settings, remains relatively unexplored and lacks established benchmarks. This paper aims to fill this gap by investigating the effectiveness of these miniature FMCW radars for robust sense-and-avoid functions on MAVs.

Methods

The core of this work lies in developing and evaluating a signal processing and avoidance pipeline for FMCW radar data on a MAV.

  • Detection: The FMCW radar transmits a saw-tooth FM carrier wave. The received signals, shifted in time due to range, are mixed with the transmitted signal to produce an intermediate frequency (IF) signal. A fast-Fourier transform (FFT) is applied to this IF signal to identify peaks corresponding to object ranges. The horizontal bearing is determined by comparing the phases from two receiver antennas.
  • Velocity Estimation: A second set of FFTs is performed over multiple chirps to extract the Doppler shift, which corresponds to the radial velocity of detected objects. This results in a 2D FFT matrix that resolves objects by both range and radial velocity.
  • Filtering and Tracking: Due to the inherent noise of the radar sensor, data association and tracking are crucial. A global nearest neighborhood (GNN) algorithm is used to calculate a cost matrix for associating detections with existing tracked objects or clutter. This is followed by a Kalman filter (KF) to predict object positions and estimate tangential velocity. The system also handles track initiation (birth) and termination (death) based on covariance thresholds.
  • Avoidance Algorithm: The paper employs the Velocity Obstacles (VO) method for obstacle avoidance. This algorithm calculates the set of MAV velocities that would lead to a collision with an object, considering the radii of both the MAV and the obstacle. It then computes a desired MAV velocity that avoids these collision cones, taking into account the MAV's own velocity limitations.

Datasets

The paper utilizes a custom dataset for evaluation:

  • Obstacle Avoidance Dataset (ODA_Dataset): This dataset was acquired in a flying arena equipped with an OptiTrack motion tracking system, providing ground truth positions for both the MAV and obstacles. It contains obstacle avoidance trials.

Key Insights and Ideas

  • Radar as a Viable Standalone Sensor: The study demonstrates that a miniature FMCW radar, despite its inherent noise, can be effectively used as a standalone sensor for reliable obstacle detection and avoidance on MAVs, even in conditions where vision-based sensors fail.
  • Importance of Robust Filtering: The noisy nature of FMCW radar necessitates sophisticated filtering and data association techniques (like GNN and KF) to extract meaningful information and enable accurate tracking.
  • Velocity Obstacles for Dynamic Avoidance: The integration of the Velocity Obstacles algorithm allows the MAV to dynamically adjust its trajectory to avoid collisions, considering the relative velocities of itself and the obstacles.
  • Trade-offs in Sensor Design: The paper highlights that while the sensor's limited number of receiver antennas restricts simultaneous detection in the same range bin, multi-frame processing can overcome this limitation, albeit with a reduced update frequency.

Key Results

  • The MAV successfully avoided obstacles (cardboard poles) when controlled by the onboard radar-based avoidance system, performing comparably to human pilots in test scenarios.
  • The system demonstrated the ability to detect and avoid collisions even when obstacles were moving faster than the MAV, though tracking further obstacles could be challenging due to sensor noise.
  • The error in bearing and range increased approximately linearly as obstacles moved towards the edge of the radar's Field of View (FOV).
  • The implemented system successfully processed radar data, performed multi-target tracking, and executed avoidance maneuvers in real-time on a MAV.

Limitations and Future Work

  • Sensor Noise: The paper acknowledges the noisy nature of the FMCW radar, which requires significant filtering and can lead to temporary tracking errors, especially during MAV trajectory changes.
  • Limited Receiver Antennas: The sensor's limitation of only two receiver antennas restricts simultaneous detection within the same range bin, impacting the update frequency for complex scenarios.
  • Single Obstacle Avoidance Focus: The avoidance algorithm, for simplicity, primarily considered the nearest obstacle, which could be a limitation in scenarios with multiple immediate threats.
  • Future Work: The authors suggest exploring 77GHz FMCW radars for improved resolution and bandwidth in cramped spaces, potentially enabling wall detection and classification. They also propose investigating more robust data association algorithms (like DBSCAN) for clustering detections and incorporating more receiver antennas for enhanced detection capabilities. Fusing radar data with other sensors (like event-based cameras) is also mentioned as a promising direction to compensate for individual sensor limitations.

Summary updated

Abstract

Micro Air Vehicles (MAVs) are increasingly being used for complex or hazardous tasks in enclosed and cluttered environments such as surveillance or search and rescue. With this comes the necessity for sensors that can operate in poor visibility conditions to facilitate with navigation and avoidance of objects or people. Radar sensors in particular can provide more robust sensing of the environment when traditional sensors such as cameras fail in the presence of dust, fog or smoke. While extensively used in autonomous driving, miniature FMCW radars on MAVs have been relatively unexplored. This study aims to investigate to what extent this sensor is of use in these environments by employing traditional signal processing such as multi-target tracking and velocity obstacles. The viability of the solution is evaluated with an implementation on board a MAV by running trial tests in an indoor environment containing obstacles and by comparison with a human pilot, demonstrating the potential for the sensor to provide a more robust sense and avoid function in fully autonomous MAVs.
BibTeX
@inproceedings{wessendorp2021obstacle,
  author = {Nikhil Wessendorp and Raoul Dinaux and Julien Dupeyroux and Guido de Croon},
  title = {{Obstacle Avoidance onboard MAVs using a FMCW RADAR}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2021},
  url = {https://openreview.net/forum?id=4OedO40RnHS},
}
Figure