[Paper Review] RMF-Owl: A Collision-Tolerant Flying Robot for Autonomous Subterranean Exploration

Paolo De Petris, Huan X. Nguyen+5 more|arXiv (Cornell University)|Feb 22, 2022

Robotic Path Planning Algorithms4 citations

TL;DR

RMF-Owl is a collision-tolerant, autonomous aerial robot designed for resilient subterranean exploration in GNSS-denied, obstacle-rich environments. It integrates robust onboard localization and mapping with a collision-tolerant airframe and real-time object detection, enabling safe, long-duration autonomous exploration of complex underground settings, as validated in field tests exceeding 200m flight distance and 8.1 minutes of flight time.

ABSTRACT

This work presents the design, hardware realization, autonomous exploration and object detection capabilities of RMF-Owl, a new collision-tolerant aerial robot tailored for resilient autonomous subterranean exploration. The system is custom built for underground exploration with focus on collision tolerance, resilient autonomy with robust localization and mapping, alongside high-performance exploration path planning in confined, obstacle-filled and topologically complex underground environments. Moreover, RMF-Owl offers the ability to search, detect and locate objects of interest which can be particularly useful in search and rescue missions. A series of results from field experiments are presented in order to demonstrate the system's ability to autonomously explore challenging unknown underground environments.

Motivation & Objective

• To develop a collision-tolerant aerial robot capable of autonomous exploration in confined, topologically complex, and GNSS-denied subterranean environments.
• To enable resilient autonomy through robust onboard localization and mapping despite sensor degradation and control inaccuracies.
• To integrate real-time object detection and localization for search and rescue applications in underground settings.
• To validate the system’s performance through field experiments in real-world subterranean environments such as mines and test galleries.

Proposed method

• The robot features a custom-designed, collision-tolerant airframe with a rigid frame and protective cage to withstand impacts during navigation in narrow, obstacle-filled passages.
• An onboard autonomy stack integrates visual-inertial SLAM for real-time localization and mapping, enabling operation in GNSS-denied environments.
• A path planning module generates collision-free trajectories based on the reconstructed map, prioritizing exploration of unvisited regions.
• A YOLO-based object detection pipeline identifies and localizes objects of interest, reporting class probabilities and sending downsampled images to the ground station.
• The system uses 5GHz WiFi for communication when a mesh network is available, sharing mapping data, odometry, battery status, and detection reports.
• Field deployments utilize a combination of onboard perception, autonomous takeoff/landing, and homing strategies to ensure safe return to the takeoff point.

Experimental results

Research questions

• RQ1 How can a flying robot achieve resilient autonomy in subterranean environments with high obstacle density and limited sensor availability?
• RQ2 What design principles enable effective collision tolerance without compromising agility and payload capacity in confined spaces?
• RQ3 Can a single robot autonomously explore and map complex underground topologies while detecting and localizing objects of interest?
• RQ4 To what extent can onboard localization and mapping systems maintain accuracy in GNSS-denied, sensor-degraded conditions?
• RQ5 How does the integration of object detection enhance the utility of autonomous exploration robots in search and rescue missions?

Key findings

• RMF-Owl successfully completed autonomous exploration missions in the Løkken Mine, covering over 200 meters with a flight time of 6.6 minutes, navigating through corridors less than 2.5 meters wide.
• In the Versuchsstollen Hagerbach Test Gallery, the robot explored a 14-meter-wide section, achieving a flight time of 8.1 minutes and a total distance exceeding 200 meters at a maximum speed of 1.2 m/s.
• The robot demonstrated robust visual-inertial SLAM performance, maintaining accurate localization and mapping in complete darkness and in environments with significant structural clutter.
• Object detection was successfully performed in-flight, with the system identifying and localizing artifacts and transmitting class probabilities and downsampled images to the ground station.
• The collision-tolerant design allowed the robot to survive multiple impacts during navigation, confirming its resilience in unpredictable subterranean conditions.
• The system achieved safe autonomous takeoff, exploration, and return-to-home in both narrow and expansive underground settings, validating its resilience and autonomy.